dissenting.

In a sensible system of criminal justice, wrongful conviction is avoided by establishing, at the trial level, lines of procedural legality that leave ample margins of safety (for example, the requirement that guilt be proved beyond a reasonable doubt) — not by providing recurrent and repetitive appellate review of whether the facts in the record show those lines to have been narrowly crossed. The defect of the latter system was described, with characteristic candor, by Justice Jackson:

“Whenever decisions of one court are reviewed by another, a percentage of them are reversed. That reflects a difference in outlook normally found between personnel comprising different courts. However, reversal by a higher court is not proof that justice is thereby better done.” Brown v. Allen, 344 U. S. 443, 540 (1953) (opinion concurring in result).

Since this Court has long shared Justice Jackson’s view, today’s opinion — which considers a fact-bound claim of error rejected by every court, state and federal, that previously heard it — is, so far as I can tell, wholly unprecedented. The Court has adhered to the policy that, when the petitioner claims only that a concededly correct view of the law was incorrectly applied to the facts, certiorari should generally (i. e., except in cases of the plainest error) be denied. United States v. Johnston, 268 U. S. 220, 227 (1925). That policy has been observed even when the fact-bound assessment of the federal court of appeals has differed from that of the district court, Sumner v. Mata, 449 U. S. 539, 543 (1981); and under what we have called the “two-court rule,” the policy has been applied with particular rigor when dis*457trict court and court of appeals are in agreement as to what conclusion the record requires. See, e. g., Graver Tank & Mfg. Co. v. Linde Air Products Co., 336 U. S. 271, 275 (1949). How much the more should the policy be honored in this case, a federal habeas proceeding where not only both lower federal courts but also the state courts on postconviction review have all reviewed and rejected precisely the fact-specific claim before us. Cf. 28 U. S. C. § 2254(d) (requiring federal habeas courts to accord a presumption of correctness to state-court findings of fact); Sumner, supra, at 550, n. 3. Instead, however, the Court not only grants certiorari to consider whether the Court of Appeals (and all the previous courts that agreed with it) was correct as to what the facts showed in a case where the answer is far from clear, but in the process of such consideration renders new findings of fact and judgments of credibility appropriate to a trial court of original jurisdiction. See, e. g., ante, at 425 (“Beanie seemed eager to cast suspicion on Kyles”); ante, at 441, n. 12 (“Record photographs of Beanie . . . depict a man possessing a medium build”); ante, at 449, n. 18 (“the record photograph of the homemade holster indicates .. .”).

The Court says that we granted certiorari “[b]ecause ‘[o]ur duty to search for constitutional error with painstaking care is never more exacting than it is in a capital case,’ Burger v. Kemp, 483 U. S. 776, 785 (1987).” Ante, at 422. The citation is perverse, for the reader who looks up the quoted opinion will discover that the very next sentence confirms the traditional practice from which the Court today glaringly departs: “Nevertheless, when the lower courts have found that [no constitutional error occurred],... deference to the shared conclusion of two reviewing courts prevents] us from substituting speculation for their considered opinions.” Burger v. Kemp, 483 U. S. 776, 785 (1987).

The greatest puzzle of today’s decision is what could have caused this capital case to be singled out for favored treatment. Perhaps it has been randomly selected as a symbol, *458to reassure America that the United States Supreme Court is reviewing capital convictions to make sure no factual error has been made. If so, it is a false symbol, for we assuredly do not do that. At, and during the week preceding, our February 24 Conference, for example, we considered and disposed of 10 petitions in capital cases, from seven States. We carefully considered whether the convictions and sentences in those cases had been obtained in reliance upon correct principles of federal law; but if we had tried to consider, in addition, whether those correct principles had been applied, not merely plausibly, but accurately, to the particular facts of each case, we would have done nothing else for the week. The reality is that responsibility for factual accuracy, in capital cases as in other cases, rests elsewhere — with trial judges and juries, state appellate courts, and the lower federal courts; we do nothing but encourage foolish reliance to pretend otherwise.

Straining to suggest a legal error in the decision below that might warrant review, the Court asserts that “[t]here is room to debate whether the two judges in the majority in the Court of Appeals made an assessment of the cumulative effect of the evidence,” ante, at 440. In support of this it quotes isolated sentences of the opinion below that supposedly “dismissed] particular items of evidence as immaterial,” ibid. This claim of legal error does not withstand minimal scrutiny. The Court of Appeals employed precisely the same legal standard that the Court does. Compare 5 F. 3d 806, 811 (CA5 1993) (“We apply the [United States v.] Bagley[, 473 U. S. 667 (1985),] standard here by examining whether it is reasonably probable that, had the undisclosed information been available to Kyles, the result would have been different”), with ante, at 441 (“In this case, disclosure of the suppressed evidence to competent counsel would have made a different result reasonably probable”). Nor did the Court of Appeals announce a rule of law, that might have precedential force in later cases, to the effect that Bagley *459requires a series of independent materiality evaluations; in fact, the court said just the contrary. See 5 F. 3d, at 817 (“[W]e are not persuaded that it is reasonably probable that the jury would have found in Kyles’ favor if exposed to any or all of the undisclosed materials”) (emphasis added). If the decision is read, shall we say, cumulatively, it is clear beyond cavil that the court assessed the cumulative effect of the Brady evidence in the context of the whole record. See 5 F. 3d, at 807 (basing its rejection of petitioner’s claim on “a complete reading of the record”); id., at 811 (“Rather than reviewing the alleged Brady materials in the abstract, we will examine the evidence presented at trial and how the extra materials would have fit”); id., at 813 (“We must bear [the eyewitness testimony] in mind while assessing the probable effect of other undisclosed information”). It is, in other words, the Court itself which errs in the manner that it accuses the Court of Appeals of erring: failing to consider the material under review as a whole. The isolated snippets it quotes from the decision merely do what the Court’s own opinion acknowledges must be done: to “evaluate the tendency and force of the undisclosed evidence item by item; there is no other way.” Ante, at 436, n. 10. Finally, the Court falls back on this: “The result reached by the Fifth Circuit majority is compatible with a series of independent materiality evaluations, rather than the cumulative evaluation required by Bagley,” ante, at 441. In other words, even though the Fifth Circuit plainly enunciated the correct legal rule, since the outcome it reached would not properly follow from that rule, the Fifth Circuit must in fact (and unbeknownst to itself) have been applying an incorrect legal rule. This effectively eliminates all distinction between mistake in law and mistake in application.

What the Court granted certiorari to review, then, is not a decision on an issue of federal law that conflicts with a decision of another federal or state court; nor even a decision announcing a rule of federal law that because of its novelty *460or importance might warrant review despite the lack of a conflict; nor yet even a decision that patently errs in its application of an old rule. What we have here is an intensely fact-specific case in which the court below unquestionably applied the correct rule of law and did not unquestionably err — precisely the type of case in which we are most inclined to deny certiorari. But despite all of that, I would not have dissented on the ground that the writ of certiorari should be dismissed as improvidently granted. Since the majority is as aware of the limits of our capacity as I am, there is little fear that the grant of certiorari in a case of this sort will often be repeated — which is to say little fear that today’s grant has any generalizable principle behind it. I am still forced to dissent, however, because, having improvidently decided to review the facts of this case, the Court goes on to get the facts wrong. Its findings are in my view clearly erroneous, ef. Fed. Rule Civ. Proc. 52(a), and the Court’s verdict would be reversed if there were somewhere further to appeal.

I

Before proceeding to detailed consideration of the evidence, a few general observations about the Court’s methodology are appropriate. It is fundamental to the discovery rule of Brady v. Maryland, 373 U. S. 83 (1963), that the materiality of a failure to disclose favorable evidence “must be evaluated in the context of the entire record.” United States v. Agurs, 427 U. S. 97, 112 (1976). It is simply not enough to show that the undisclosed evidence would have allowed the defense to weaken, or even to “destrofy],” ante, at 441, the particular prosecution witnesses or items of prosecution evidence to which the undisclosed evidence relates. It is petitioner’s burden to show that in light of all the evidence, including that untainted by the Brady violation, it is reasonably probable that a jury would have entertained a reasonable doubt regarding petitioner’s guilt. See United States v. Bagley, 473 U. S. 667, 682 (1985); Agurs, *461supra, at 112-113. The Court’s opinion fails almost entirely to take this principle into account. Having spent many pages assessing the effect of the Brady material on two prosecution witnesses and a few items of prosecution evidence, ante, at 441-451, it dismisses the remainder of the evidence against Kyles in a quick page-and-a-half, ante, at 451-453. This partiality is confirmed in the Court’s attempt to “recap .. .the suppressed evidence and its significance for the prosecution,” ante, at 453 (emphasis added), which omits the required comparison between that evidence and the evidence that was disclosed. My discussion of the record will present the half of the analysis that the Court omits, emphasizing the evidence concededly unaffected by the Brady violation which demonstrates the immateriality of the violation.

In any analysis of this case, the desperate implausibility of the theory that petitioner put before the jury must be kept firmly in mind. The first half of that theory — designed to neutralize the physical evidence (Mrs. Dye’s purse in his garbage, the murder weapon behind his stove) — was that petitioner was the victim of a “frame-up” by the police informer and evil genius, Beanie. Now it is not unusual for a guilty person who knows that he is suspected of a crime to try to shift blame to someone else; and it is less common, but not unheard of, for a guilty person who is neither suspected nor subject to suspicion (because he has established a perfect alibi), to call attention to himself by coming forward to point the finger at an innocent person. But petitioner’s theory is that the guilty Beanie, who could plausibly be accused of the crime (as petitioner’s brief amply demonstrates), but who was not a suspect any more than Kyles was (the police as yet had no leads, see ante, at 424), injected both Kyles and himself into the investigation in order to get the innocent Kyles convicted.1 If this were not stupid enough, the *462wicked Beanie is supposed to have suggested that the police search his victim’s premises a full day before he got around to planting the incriminating evidence on the premises.

The second half of petitioner’s theory was that he was the victim of a quadruple coincidence, in which four eyewitnesses to the crime mistakenly identified him as the murderer — three picking him out of a photo array without hesitation, and all four affirming their identification in open court after comparing him with Beanie. The extraordinary mistake petitioner had to persuade the jury these four witnesses made was not simply to mistake the real killer, Beanie, for the very same innocent third party (hard enough to believe), but in addition to mistake him for the very man Beanie had chosen to frame — the last and most incredible level of coincidence. However small the chance that the jury would believe any one of those improbable scenarios, the likelihood that it would believe them all together is far smaller. The Court concludes that it is “reasonably probable” the undisclosed witness interviews would have persuaded the jury of petitioner’s implausible theory of mistaken eyewitness testimony, and then argues that it is “reasonably probable” the undisclosed information regarding Beanie would have persuaded the jury of petitioner’s implausible theory regarding the incriminating physical evidence. I think neither of those conclusions is remotely true, but even if they were the Court would still be guilty of a fallacy in declaring victory on each implausibility in turn, and thus victory on the whole, *463without considering the infinitesimal probability of the jury’s swallowing the entire concoction of implausibility squared.

This basic error of approaching the evidence piecemeal is also what accounts for the Court’s obsessive focus on the credibility or culpability of Beanie, who did not even testify at trial and whose credibility or innocence the State has never once avowed. The Court’s opinion reads as if either petitioner or Beanie must be telling the truth, and any evidence tending to inculpate or undermine the credibility of the one would exculpate or enhance the credibility of the other. But the jury verdict in this case said only that petitioner was guilty of the murder. That is perfectly consistent with the possibilities that Beanie repeatedly lied, ante, at 445, that he was an accessory after the fact, cf. ante, at 445-446, or even that he planted evidence against petitioner, ante, at 448. Even if the undisclosed evidence would have allowed the defense to thoroughly impeach Beanie and to suggest the above possibilities, the jury could well have believed all of those things and yet have condemned petitioner because it could not believe that all four of the eyewitnesses were similarly mistaken.2

Of course even that much rests on the premise that competent counsel would run the terrible risk of calling Beanie, a witness whose “testimony almost certainly would have inculpated [petitioner]” and whom “any reasonable attorney would perceive ... as a ‘loose cannon.’” 5 F. 3d, at 818. Perhaps because that premise seems so implausible, the Court retreats to the possibility that petitioner’s counsel, *464even if not calling Beanie to the stand, could have used the evidence relating to Beanie to attack “the reliability of the investigation.” Ante, at 446. But that is distinctly less effective than substantive evidence bearing on the guilt or innocence of the accused. In evaluating Brady claims, we assume jury conduct that is both rational and obedient to the law. We do not assume that even though the whole mass of the evidence, both disclosed and undisclosed, shows petitioner guilty beyond a reasonable doubt, the jury will punish sloppy investigative techniques by setting the defendant free. Neither Beanie nor the police were on trial in this case. Petitioner was, and no amount of collateral evidence could have enabled his counsel to move the mountain of direct evidence against him.

II

The undisclosed evidence does not create a “'reasonable probability’ of a different result.” Ante, at 434 (quoting United States v. Bagley, 473 U. S., at 682). To begin with the eyewitness testimony: Petitioner’s basic theory at trial was that the State’s four eyewitnesses happened to mistake Beanie, the real killer, for petitioner, the man whom Beanie was simultaneously trying to frame. Police officers testified to the jury, and petitioner has never disputed, that three of the four eyewitnesses (Territo, Smallwood, and Williams) were shown a photo lineup of six young men four days after the shooting and, without aid or duress, identified petitioner as the murderer; and that all of them, plus the fourth eyewitness, Kersh, reaffirmed their identifications at trial after petitioner and Beanie were made to stand side by side.

Territo, the first eyewitness called by the State, was waiting at a red light in a truck 30 or 40 yards from the Schwegmann’s parking lot. He saw petitioner shoot Mrs. Dye, start her car, drive out onto the road, and pull up just behind Territo’s truck. When the light turned green petitioner pulled *465beside Territo and stopped while waiting to make a turn. Petitioner looked Territo full in the face. Territo testified, “I got a good look at him. If I had been in the passenger seat of the little truck, I could have reached out and not even stretched my arm out, I could have grabbed hold of him.” Tr. 13-14 (Dec. 6, 1984). Territo also testified that a detective had shown him a picture of Beanie and asked him if the picture “could have been the guy that did it. I told him no.” Id., at 24. The second eyewitness, Kersh, also saw petitioner shoot Mrs. Dye. When asked whether she got “a good look” at him as he drove away, she answered “yes.” Id., at 32. She also answered “yes” to the question whether she “got to see the side of his face,” id., at 31, and said that while petitioner was stopped she had driven to within reaching distance of the driver’s-side door of Mrs. Dye’s ear and stopped there. Id., at 34. The third eyewitness, Small-wood, testified that he saw petitioner shoot Mrs. Dye, walk to the car, and drive away. Id., at 42. Petitioner drove slowly by, within a distance of 15 or 25 feet, id., at 43-45, and Smallwood saw his face from the side. Id., at 43. The fourth eyewitness, Williams, who had been working outside the parking lot, testified that “the gentleman came up the side of the car,” struggled with Mrs. Dye, shot her, walked around to the driver’s side of the car, and drove away. Id., at 52. Williams not only “saw him before he shot her,” id., at 54, but watched petitioner drive slowly by “within less than ten feet.” Ibid. When asked “[d]id you get an opportunity to look at him good?”, Williams said, “I did.” Id., at 55.

The Court attempts to dispose of this direct, unqualified, and consistent eyewitness testimony in two ways. First, by relying on a theory so implausible that it was apparently not suggested by petitioner’s counsel until the oral-argument-cum-evidentiary-hearing held before us, perhaps because it is a theory that only the most removed appellate court could *466love. This theory is that there is a reasonable probability that the jury would have changed its mind about the eyewitness identification because the Brady material would have permitted the defense to argue that the eyewitnesses only got a good look at the killer when he was sitting in Mrs. Dye’s car, and thus could identify him, not by his height and build, but only by his face. Never mind, for the moment, that this is factually false, since the Brady material showed that only one of the four eyewitnesses, Smallwood, did not see the killer outside the car.3 And never mind, also, the dubious premise that the build of a man 6-feet tall (like petitioner) is indistinguishable, when seated behind the wheel, from that of a man less than 5V2-feet tall (like Beanie). To assert that unhesitant and categorical identification by four witnesses who viewed the killer, close-up and with the sun high in the sky, would not eliminate reasonable doubt if it were based only on facial characteristics, and not on height and build, is quite simply absurd. Facial features are the primary means by which human beings recognize one another. That is why police departments distribute “mug” shots of wanted felons, rather than Ivy-League-type posture pictures; it is why bank robbers wear stockings over their faces instead of floor-length capes over their shoulders; it is why the Lone Ranger wears a mask instead of a poncho; and it is why a criminal defense lawyer who seeks to destroy an *467identifying witness by asking “You admit that you saw only the killer’s face?” will be laughed out of the courtroom.

It would be different, of course, if there were evidence that Kyles’s and Beanie’s faces looked like twins, or at least bore an unusual degree of resemblance. That facial resemblance would explain why, if Beanie committed the crime, all four witnesses picked out Kyles at first (though not why they continued to pick him out when he and Beanie stood side-by-side in court), and would render their failure to observe the height and build of the killer relevant. But without evidence of facial similarity, the question “You admit that you saw only the killer’s face?” draws no blood; it does not explain any witness’s identification of petitioner as the killer. While the assumption of facial resemblance between Kyles and Beanie underlies all of the Court’s repeated references to the partial concealment of the killer’s body from view, see, e. g., ante, at 442-443, 443-444, n. 14, 445, the Court never actually says that such resemblance exists. That is because there is not the slightest basis for such a statement in the record. No court has found that Kyles and Beanie bear any facial resemblance. In fact, quite the opposite: every federal and state court that has reviewed the record photographs, or seen the two men, has found that they do not resemble each other in any respect. See 5 F. 3d, at 813 (“Comparing photographs of Kyles and Beanie, it is evident that the former is taller, thinner, and has a narrower face”); App. 181 (District Court opinion) (“The court examined all of the pictures used in the photographic line-up and compared Kyles’ and Beanie’s pictures; it finds that they did not resemble one another”); id., at 36 (state trial court findings on postconviction review) (“[Beanie] clearly and distinctly did not resemble the defendant in this case”) (emphasis in original). The District Court’s finding controls because it is not clearly erroneous, Fed. Rule Civ. Proc. 52(a), and the state court’s finding, because fairly supported by the record, must be presumed correct on habeas review. See 28 U. S. C. § 2254(d).

*468The Court’s second means of seeking to neutralize the impressive and unanimous eyewitness testimony uses the same “build-is-everything” theory to exaggerate the effect of the State’s failure to disclose the contemporaneous statement of Henry Williams. That statement would assuredly have permitted a sharp cross-examination, since it contained estimations of height and weight that fit Beanie better than petitioner. Ante, at 441-442. But I think it is hyperbole to say that the statement would have “substantially reduced or destroyed” the value of Williams’ testimony. Ante, at 441. Williams saw the murderer drive slowly by less than 10 feet away, Tr. 54 (Dec. 6, 1984), and unhesitatingly picked him out of the photo lineup. The jury might well choose to give greater credence to the simple fact of identification than to the difficult estimation of height and weight.

The Court spends considerable time, see ante, at 443, showing how Smallwood’s testimony could have been discredited to such a degree as to “rais[e] a substantial implication that the prosecutor had coached him to give it.” Ibid. Perhaps so, but that is all irrelevant to this appeal, since all of that impeaching material (except the “facial identification” point I have discussed above) was available to the defense independently of the Brady material. See ante, at 443-444, n. 14. In sum, the undisclosed statements, credited with everything they could possibly have provided to the defense, leave two prosecution witnesses (Territo and Kersh) totally untouched; one prosecution witness (Smallwood) barely affected (he saw “only” the killer’s face); and one prosecution witness (Williams) somewhat impaired (his description of the killer’s height and weight did not match Kyles). We must keep all this in due perspective, remembering that the relevant question in the materiality inquiry is not how many points the defense could have scored off the prosecution witnesses, but whether it is reasonably probable that the new evidence would have caused the jury to accept the basic thesis that all four witnesses were mistaken. I think it plainly *469is not. No witness involved in the case ever identified anyone but petitioner as the murderer. Their views of the crime and the escaping criminal were obtained in bright daylight from close at hand; and their identifications were reaffirmed before the jury. After the side-by-side comparison between Beanie and Kyles, the jury heard Territo say that there was “[n]o doubt in my mind” that petitioner was the murderer, Tr. 378 (Dec. 7, 1984); heard Kersh say “I know it was him.... I seen his face and I know the color of his skin. I know it. I know it’s him,” id., at 383; heard Smallwood say “I’m positive ... [b]ecause that’s the man who I seen kill that woman,” id., at 387; and heard Williams say “[n]o doubt in my mind,” id., at 391. With or without the Brady evidence, there could be no doubt in the mind of the jury either.

There remains the argument that is the major contribution of today’s opinion to Brady litigation; with our endorsement, it will surely be trolled past appellate courts in all future failure-to-disclose cases. The Court argues that “the effective impeachment of one eyewitness can call for a new trial even though the attack does not extend directly to others, as we have said before.” Ante, at 445 (citing Agurs v. United States, 427 U. S., at 112-113, n. 21). It would be startling if we had “said [this] before,” since it assumes irrational jury conduct. The weakening of one witness’s testimony does not weaken the unconnected testimony of another witness; and to entertain the possibility that the jury will give it such an effect is incompatible with the whole idea of a materiality standard, which presumes that the incriminating evidence that would have been destroyed by proper disclosure can be logically separated from the incriminating evidence that would have remained unaffected. In fact we have said nothing like what the Court suggests. The opinion’s only authority for its theory, the cited footnote from Agurs, was appended to the proposition that “[a Brady] omission must be evaluated in the context of the entire record,” 427 U. S., *470at 112. In accordance with that proposition, the footnote recited a hypothetical that shows how a witness’s testimony could have been destroyed by withheld evidence that contradicts the witness.4 That is worlds apart from having it destroyed by the corrosive effect of withheld evidence that impeaches (or, as here, merely weakens) some other corroborating witness.

The physical evidence confirms the immateriality of the nondisclosures. In a garbage bag outside petitioner’s home the police found Mrs. Dye’s purse and other belongings. Inside his home they found, behind the kitchen stove, the .32-caliber revolver used to kill Mrs. Dye; hanging in a wardrobe, a homemade shoulder holster that was “a perfect fit” for the revolver, Tr. 74 (Dec. 6, 1984) (Detective Dillman); in a dresser drawer in the bedroom, two boxes of gun cartridges, one containing only .32-caliber rounds of the same brand found in the murder weapon, another containing .22, .32, and .38-caliber rounds; in a kitchen cabinet, eight empty Schwegmann’s bags; and in a cupboard underneath that cabinet, one Schwegmann’s bag containing 15 cans of pet food. Petitioner’s account at trial was that Beanie planted the purse, gun, and holster, that petitioner received the ammunition from Beanie as collateral for a loan, and that petitioner had bought the pet food the day of the murder. That account strains credulity to the breaking point.

*471The Court is correct that the Brady material would have supported the claim that Beanie planted Mrs. Dye’s belongings in petitioner’s garbage and (to a lesser degree) that Beanie planted the gun behind petitioner’s stove. Ante, at 448. But we must see the whole story that petitioner presented to the jury. Petitioner would have it that Beanie did not plant the incriminating evidence until the day after he incited the police to search petitioner’s home. Moreover, he succeeded in surreptitiously placing the gun behind the stove, and the matching shoulder holster in the wardrobe, while at least 10 and as many as 19 people were present in petitioner’s small apartment.5 Beanie, who was wearing blue jeans and either a “tank-top” shirt, Tr. 302 (Dec. 7,1984) (Cathora Brown), or a short-sleeved shirt, id., at 351 (petitioner), would have had to be concealing about his person not only the shoulder holster and the murder weapon, but also a different gun with tape wrapped around the barrel that he showed to petitioner. Id., at 352. Only appellate judges could swallow such a tale. Petitioner’s only supporting evidence was Johnny Burns’s testimony that he saw Beanie stooping behind the stove, presumably to plant the gun. Id., at 262-263. Burns’s credibility on the stand can perhaps best be gauged by observing that the state judge who presided over petitioner’s trial stated, in a postconviction proceeding, that “[I] ha[ve] chosen to totally disregard everything that [Burns] has said,” App. 35. See also id., at 165 (District Court opinion) (“Having reviewed the entire record, this court without hesitation concurs with the trial court’s determination concerning the credibility of [Burns]”). Burns, by the way, who repeatedly stated at trial that Beanie was his “best friend,” Tr. 279 (Dec. 7, 1984), has since been *472tried and convicted for killing Beanie. See State v. Burnes, 533 So. 2d 1029 (La. App. 1988).6

Petitioner did not claim that the ammunition had been planted. The police found a .22-caliber rifle under petitioner’s mattress and two boxes of ammunition, one containing .22, .32, and .38-caliber rounds, another containing only .32-caliber rounds of the same brand as those found loaded in the murder weapon. Petitioner’s story was that Beanie gave him the rifle and the .32-caliber shells as security for a loan, but that he had taken the .22-caliber shells out of the box. Tr. 353, 355 (Dec. 7, 1984). Put aside that the latter detail was contradicted by the facts; but consider the inherent implausibility of Beanie’s giving petitioner collateral in the form of a box containing only .32 shells, if it were true that petitioner did not own a .32-caliber gun. As the Fifth Circuit wrote, “[t]he more likely inference, apparently chosen by the jury, is that [petitioner] possessed .32-caliber ammunition because he possessed a .32-caliber firearm.” 5 F. 3d, at 817.

We come to the evidence of the pet food, so mundane and yet so very damning. Petitioner’s confused and changing explanations for the presence of 15 cans of pet food in a Schwegmann’s bag under the sink must have fatally undermined his credibility before the jury. See App. 36 (trial judge finds that petitioner’s “obvious lie” concerning the pet food “may have been a crucial bit of evidence in the minds of the jurors which caused them to discount the entire de*473fense in this case”). The Court disposes of the pet food evidence as follows:

“The fact that pet food was found in Kyles’s apartment was consistent with the testimony of several defense witnesses that Kyles owned a dog and that his children fed stray cats. The brands of pet food found were only two of the brands that Dye typically bought, and these two were common, whereas the one specialty brand that was found in Dye’s apartment after her murder, Tr. 180 (Dec. 7, 1984), was not found in Kyles’s apartment, id., at 188. Although Kyles was wrong in describing the cat food as being on sale the day he said he bought it, he was right in describing the way it was priced at Schwegmann’s market, where he commonly shopped.” Ante, at 451-452; see also ante, at 452, n. 20.

The full story is this. Mr. and Mrs. Dye owned two cats and a dog, Tr. 178 (Dec. 7, 1984), for which she regularly bought varying brands of pet food, several different brands at a time. Id., at 179,180. Found in Mrs. Dye’s home after her murder were the brands Nine Lives, Kalkan, and Puss n’ Boots. Id., at 180. Found in petitioner’s home were eight cans of Nine Lives, four cans of Kalkan, and three cans of Cozy Kitten. Id., at 188. Since we know that Mrs. Dye had been shopping that day and that the murderer made off with her goods, petitioner’s possession of these items was powerful evidence that he was the murderer. Assuredly the jury drew that obvious inference. Pressed to explain why he just happened to buy 15 cans of pet food that very day (keep in mind that petitioner was a very poor man, see id., at 329, who supported a common-law wife, a mistress, and four children), petitioner gave the reason that “it was on sale.” Id., at 341. The State, however, introduced testimony from the Schwegmann’s advertising director that the pet food was not on sale that day. Id., at 395. The dissenting judge below tried to rehabilitate petitioner’s testimony *474by interpreting the “on sale” claim as meaning “for sale,” a reference to the pricing of the pet food (e. g., “3 for 89 cents”), which petitioner claimed to have read on a shelf sign in the store. Id., at 343. But unless petitioner was parodying George Leigh Mallory, “because it was for sale” would have been an irrational response to the question it was given in answer to: Why did you buy so many cans? In any event, the Schwegmann’s employee also testified that store policy was not to put signs on the shelves at all. Id., at 398-399. The sum of it is that petitioner, far from explaining the presence of the pet food, doubled the force of the State’s evidence by perjuring himself before the jury, as the state trial judge observed. See supra, at 472-473.7

I will not address the list of cars in the Schwegmann’s parking lot and the receipt, found in the victim’s car, that bore petitioner’s fingerprints. These were collateral matters that provided little evidence of either guilt or innocence. The list of cars, which did not contain petitioner’s automobile, would only have served to rebut the State’s introduction of a photograph purporting to show petitioner’s car in the parking lot; but petitioner does not contest that the list was not comprehensive, and that the photograph was taken about six hours before the list was compiled. See 5 F. 3d, at 816. *475Thus its rebuttal value would have been marginal at best. The receipt — although it showed that petitioner must at some point have been both in Schwegmann’s and in the murdered woman’s car — was as consistent with petitioner’s story as with the State’s. See ante, at 452.

* * *

The State presented to the jury a massive core of evidence (including four eyewitnesses) showing that petitioner was guilty of murder, and that he lied about his guilt. The effect that the Brady materials would have had in chipping away at the edges of the State’s case can only be called immaterial. For the same reasons I reject petitioner’s claim that the Brady materials would have created a “residual doubt” sufficient to cause the sentencing jury to withhold capital punishment.

I respectfully dissent.

delivered the opinion of the Court.

After his first trial in 1984 ended in a hung jury, petitioner Curtis Lee Kyles was tried again, convicted of first-degree murder, and sentenced to death. On habeas review, we follow the established rule that the state’s obligation under Brady v. Maryland, 373 U. S. 83 (1963), to disclose evidence favorable to the defense, turns on the cumulative effect of all such evidence suppressed by the government, and we hold that the prosecutor remains responsible for gauging that effect regardless of any failure by the police to bring favorable evidence to the prosecutor’s attention. Because the net effect of the evidence withheld by the State in this case raises *422a reasonable probability that its disclosure would have produced a different result, Kyles is entitled to a new trial.

I

Following the mistrial when the jury was unable to reach a verdict, Kyles’s subsequent conviction and sentence of death were affirmed on direct appeal. State v. Kyles, 513 So. 2d 265 (La. 1987), cert. denied, 486 U. S. 1027 (1988). On state collateral review, the trial court denied relief, but the Supreme Court of Louisiana remanded for an evidentiary hearing on Kyles’s claims of newly discovered evidence. During this state-court proceeding, the defense was first able to present certain evidence, favorable to Kyles, that the State had failed to disclose before or during trial. The state trial court nevertheless denied relief, and the State Supreme Court denied Kyles’s application for discretionary review. State ex rel. Kyles v. Butler, 566 So. 2d 386 (La. 1990).

Kyles then filed a petition for habeas corpus in the United States District Court for the Eastern District of Louisiana, which denied the petition. The Court of Appeals for the Fifth Circuit affirmed by a divided vote. 5 F. 3d 806 (1993). As we explain, infra, at 440-441, there is reason to question whether the Court of Appeals evaluated the significance of undisclosed evidence under the correct standard. Because “[o]ur duty to search for constitutional error with painstaking care is never more exacting than it is in a capital case,” Burger v. Kemp, 483 U. S. 776, 785 (1987),1 we granted certiorari, 511 U. S. 1051 (1994), and now reverse.

*423II

A

The record indicates that, at about 2:20 p.m. on Thursday, September 20, 1984, 60-year-old Dolores Dye left the Schwegmann Brothers’ store (Schwegmann’s) on Old Gentilly Road in New Orleans after doing some food shopping. As she put her grocery bags into the trunk of her red Ford LTD, a man accosted her and after a short struggle drew a revolver, fired into her left temple, and killed her. The gunman took Dye’s keys and drove away in the LTD.

New Orleans police took statements from six eyewitnesses,2 who offered various descriptions of the gunman. They agreed that he was a black man, and four of them said that he had braided hair. The witnesses differed significantly, however, in their descriptions of height, age, weight, build, and hair length. Two reported seeing a man of 17 or 18, while another described the gunman as looking as old as 28. One witness described him as 5'4" or 5'5", medium build, 140-150 pounds; another described the man as slim and close to six feet. One witness said he had a mustache; none of the others spoke of any facial hair at all. One witness said the murderer had shoulder-length hair; another described the hair as “short.”

Since the police believed the killer might have driven his own car to Schwegmann’s and left it there when he drove off in Dye’s LTD, they recorded the license numbers of the cars remaining in the parking lots around the store at 9:15 p.m. on the evening of the murder. Matching these numbers with registration records produced the names and addresses of the owners of the cars, with a notation of any owner’s police *424record. Despite this list and the eyewitness descriptions, the police had no lead to the gunman until the Saturday evening after the shooting.

At 5:30 p.m., on September 22, a man identifying himself as James Joseph called the police and reported that on the day of the murder he had bought a red Thunderbird from a friend named Curtis, whom he later identified as petitioner, Curtis Kyles. He said that he had subsequently read about Dye’s murder in the newspapers and feared that the car he purchased was the victim’s. He agreed to meet with the police.

A few hours later, the informant met New Orleans Detective John Miller, who was wired with a hidden body microphone, through which the ensuing conversation was recorded. See App. 221-257 (transcript). The informant now said his name was Joseph Banks and that he was called Beanie. His actual name was Joseph Wallace.3

His story, as well as his name, had changed since his earlier call. In place of his original account of buying a Thunderbird from Kyles on Thursday, Beanie told Miller that he had not seen Kyles at all on Thursday, id., at 249-250, and had bought a red LTD the previous day, Friday, id., at 221-222, 225. Beanie led Miller to the parking lot of a nearby bar, where he had left the red LTD, later identified as Dye’s.

Beanie told Miller that he lived with Kyles’s brother-in-law (later identified as Johnny Burns),4 whom Beanie repeatedly called his “partner.” Id., at 221. Beanie described Kyles as slim, about 6-feet tall, 24 or 25 years old, with a “bush” hairstyle. Id., at 226, 252. When asked if Kyles ever wore *425his hair in plaits, Beanie said that he did but that he “had a bush” when Beanie bought the car. Id., at 249.

During the conversation, Beanie repeatedly expressed concern that he might himself be a suspect in the murder. He explained that he had been seen driving Dye’s car on Friday evening in the French Quarter, admitted that he had changed its license plates, and worried that he “could have been charged” with the murder on the basis of his possession of the LTD. Id., at 231, 246, 250. He asked if he would be put in jail. Id., at 235, 246. Miller acknowledged that Beanie’s possession of the car would have looked suspicious, id., at 247, but reassured him that he “didn’t do anything wrong,” id., at 235.

Beanie seemed eager to cast suspicion on Kyles, who allegedly made his living by “robbing people,” and had tried to kill Beanie at some prior time. Id., at 228, 245, 251. Beanie said that Kyles regularly carried two pistols, a .38 and a .32, and that if the police could “set him up good,” they could “get that same gun” used to kill Dye. Id., at 228-229. Beanie rode with Miller and Miller’s supervisor, Sgt. James Eaton, in an unmarked squad car to Desire Street, where he pointed out the building containing Kyles’s apartment. Id., at 244-246.

Beanie told the officers that after he bought the car, he and his “partner” (Burns) drove Kyles to Schwegmann’s about 9 p.m. on Friday evening to pick up Kyles’s car, described as an orange four-door Ford.5 Id., at 221, 223, 231-232, 242. When asked where Kyles’s car had been parked, Beanie replied that it had been “[o]n the same side [of the lot] where the woman was killed at.” Id., at 231. The officers later drove Beanie to Schwegmann’s, where he indicated the space where he claimed Kyles’s car had been parked. Beanie went on to say that when he and Burns had brought Kyles to pick *426up the car, Kyles had gone to some nearby bushes to retrieve a brown purse, id., at 253-255, which Kyles subsequently hid in a wardrobe at his apartment. Beanie said that Kyles had “a lot of groceries” in Schwegmann’s bags and a new baby’s potty “in the car.” Id., at 254-255. Beanie told Eaton that Kyles’s garbage would go out the next day and that if Kyles was “smart” he would “put [the purse] in [the] garbage.” Id., at 257. Beanie made it clear that he expected some reward for his help, saying at one point that he was not “doing all of this for nothing.” Id., at 246. The police repeatedly assured Beanie that he would not lose the $400 he paid for the car. Id., at 243, 246.

After the visit to Schwegmann’s, Eaton and Miller took Beanie to a police station where Miller interviewed him again on the record, which was transcribed and signed by Beanie, using his alias “Joseph Banks.” See id., at 214-220. This statement, Beanie’s third (the telephone call being the first, then the recorded conversation), repeats some of the essentials of the second one: that Beanie had purchased a red Ford LTD from Kyles for $400 on Friday evening; that Kyles had his hair “combed out” at the time of the sale; and that Kyles carried a .32 and a .38 with him “all the time.”

Portions of the third statement, however, embellished or contradicted Beanie’s preceding story and were even internally inconsistent. Beanie reported that after the sale, he and Kyles unloaded Schwegmann’s grocery bags from the trunk and back seat of the LTD and placed them in Kyles’s own car. Beanie said that Kyles took a brown purse from the front seat of the LTD and that they then drove in separate cars to Kyles’s apartment, where they unloaded the groceries. Id., at 216-217. Beanie also claimed that, a few hours later, he and his “partner” Burns went with Kyles to Schwegmann’s, where they recovered Kyles’s car and a “big brown pocket book” from “next to a building.” Id., at 218. Beanie did not explain how Kyles could have picked up his car and recovered the purse at Schwegmann’s, after Beanie *427had seen Kyles with both just a few hours earlier. The police neither noted the inconsistencies nor questioned Beanie about them.

Although the police did not thereafter put Kyles under surveillance, Tr. 94 (Dec. 6, 1984), they learned about events at his apartment from Beanie, who went there twice on Sunday. According to a fourth statement by Beanie, this one given to the chief prosecutor in November (between the first and second trials), he first went to the apartment about 2 p.m., after a telephone conversation with a police officer who asked whether Kyles had the gun that was used to kill Dye. Beanie stayed in Kyles’s apartment until about 5 p.m., when he left to call Detective John Miller. Then he returned about 7 p.m. and stayed until about 9:30 p.m., when he left to meet Miller, who also asked about the gun. According to this fourth statement, Beanie “rode around” with Miller until 3 a.m. on Monday, September 24. Sometime during those same early morning hours, detectives were sent at Sgt. Eaton’s behest to pick up the rubbish outside Kyles’s building. As Sgt. Eaton wrote in an interoffice memorandum, he had “reason to believe the victims [sic] personal papers and the Schwegmann’s bags will be in the trash.” Record, Defendant’s Exh. 17.

At 10:40 a.m., Kyles was arrested as he left the apartment, which was then searched under a warrant. Behind the kitchen stove, the police found a .32-caliber revolver containing five live rounds and one spent cartridge. Ballistics tests later showed that this pistol was used to murder Dye. In a wardrobe in a hallway leading to the kitchen, the officers found a homemade shoulder holster that fit the murder weapon. In a bedroom dresser drawer, they discovered two boxes of ammunition, one containing several .32-caliber rounds of the same brand as those found in the pistol. Back in the kitchen, various cans of cat and dog food, some of them of the brands Dye typically purchased, were found in Schwegmann’s sacks. No other groceries were identified as *428possibly being Dye’s, and no potty was found. Later that afternoon at the police station, police opened the rubbish bags and found the victim’s purse, identification, and other personal belongings wrapped in a Schwegmann’s sack.

The gun, the LTD, the purse, and the cans of pet food were dusted for fingerprints. The gun had been wiped clean. Several prints were found on the purse and on the LTD, but none was identified as Kyles’s. Dye’s prints were not found on any of the cans of pet food. Kyles's prints were found, however, on a small piece of paper taken from the front passenger-side floorboard of the LTD. The crime laboratory recorded the paper as a Schwegmann’s sales slip, but without noting what had been printed on it, which was obliterated in the chemical process of lifting the fingerprints. A second Schwegmann’s receipt was found in the trunk of the LTD, but Kyles’s prints were not found on it. Beanie’s fingerprints were not compared to any of the fingerprints found. Tr. 97 (Dec. 6, 1984).

The lead detective on the case, John Dillman, put together a photo lineup that included a photograph of Kyles (but not of Beanie) and showed the array to five of the six eyewitnesses who had given statements. Three of them picked the photograph of Kyles; the other two could not confidently identify Kyles as Dye’s assailant.

B

Kyles was indicted for first-degree murder. Before trial, his counsel filed a lengthy motion for disclosure by the State of any exculpatory or impeachment evidence. The prosecution responded that there was “no exculpatory evidence of any nature,” despite the government’s knowledge of the following evidentiary items: (1) the six contemporaneous eyewitness statements taken by police following the murder; (2) records of Beanie’s initial call to the police; (3) the tape recording of the Saturday conversation between Beanie and officers Eaton and Miller; (4) the typed and signed statement *429given by Beanie on Sunday morning; (5) the computer printout of license numbers of cars parked at Schwegmann’s on the night of the murder, which did not list the number of Kyles’s car; (6) the internal police memorandum calling for the seizure of the rubbish after Beanie had suggested that the purse might be found there; and (7) evidence linking Beanie to other crimes at Schwegmann’s and to the unrelated murder of one Patricia Leidenheimer, committed in January before the Dye murder.

At the first trial, in November, the heart of the State’s case was eyewitness testimony from four people who were at the scene of the crime (three of whom had previously picked Kyles from the photo lineup). Kyles maintained his innocence, offered supporting witnesses, and supplied an alibi that he had been picking up his children from school at the time of the murder. The theory of the defense was that Kyles had been framed by Beanie, who had planted evidence in Kyles’s apartment and his rubbish for the purposes of shifting suspicion away from himself, removing an impediment to romance with Pinky Burns, and obtaining reward money. Beanie did not testify as a witness for either the defense or the prosecution.

Because the State withheld evidence, its case was much stronger, and the defense case much weaker, than the full facts would have suggested. Even so, after four hours of deliberation, the jury became deadlocked on the issue of guilt, and a mistrial was declared.

After the mistrial, the chief trial prosecutor, Cliff Strider, interviewed Beanie. See App. 258-262 (notes of interview). Strider’s notes show that Beanie again changed important elements of his story. He said that he went with Kyles to retrieve Kyles’s car from the Schwegmann’s lot on Thursday, the day of the murder, at some time between 5 and 7:30 p.m., not on Friday, at 9 p.m., as he had said in his second and third statements. (Indeed, in his second statement, Beanie said that he had not seen Kyles at all on Thursday. Id., at *430249-250.) He also said, for the first time, that when they had picked up the car they were accompanied not only by Johnny Burns but also by Kevin Black, who had testified for the defense at the first trial. Beanie now claimed that after getting Kyles’s car they went to Black's house, retrieved a number of bags of groceries, a child’s potty, and a brown purse, all of which they took to Kyles’s apartment. Beanie also stated that on the Sunday after the murder he had been at Kyles’s apartment two separate times. Notwithstanding the many inconsistencies and variations among Beanie’s statements, neither Strider’s notes nor any of the other notes and transcripts were given to the defense.

In December 1984, Kyles was tried a second time. Again, the heart of the State’s case was the testimony of four eyewitnesses who positively identified Kyles in front of the jury. The prosecution also offered a blown-up photograph taken at the crime scene soon after the murder, on the basis of which the prosecutors argued that a seemingly two-toned car in the background of the photograph was Kyles’s. They repeatedly suggested during cross-examination of defense witnesses that Kyles had left his own car at Schwegmann’s on the day of the murder and had retrieved it later, a theory for which they offered no evidence beyond the blown-up photograph. Once again, Beanie did not testify.

As in the first trial, the defense contended that the eyewitnesses were mistaken. Kyles’s counsel called several individuals, including Kevin Black, who testified to seeing Beanie, with his hair in plaits, driving a red car similar to the victim’s about an hour after the killing. Tr. 209 (Dec. 7, 1984). Another witness testified that Beanie, with his hair in braids, had tried to sell him the car on Thursday evening, shortly after the murder. Id., at 234-235. Another witness testified that Beanie, with his hair in a “Jheri curl,” had attempted to sell him the car on Friday. Id., at 249-251. One witness, Beanie’s “partner,” Burns, testified that he had seen Beanie on Sunday at Kyles’s apartment, stooping down near *431the stove where the gun was eventually found, and the defense presented testimony that Beanie was romantically interested in Pinky Burns. To explain the pet food found in Kyles’s apartment, there was testimony that Kyles’s family kept a dog and cat and often fed stray animals in the neighborhood.

Finally, Kyles again took the stand. Denying any involvement in the shooting, he explained his fingerprints on the cash register receipt found in Dye’s car by saying that Beanie had picked him up in a red car on Friday, September 21, and had taken him to Schwegmann’s, where he purchased transmission fluid and a pack of cigarettes. He suggested that the receipt may have fallen from the bag when he removed the cigarettes.

On rebuttal, the prosecutor had Beanie brought into the courtroom. All of the testifying eyewitnesses, after viewing Beanie standing next to Kyles, reaffirmed their previous identifications of Kyles as the murderer. Kyles was convicted of first-degree murder and sentenced to death. Beanie received a total of $1,600 in reward money. See Tr. of Hearing on Post-Conviction Relief 19-20 (Feb. 24, 1989); id., at 114 (Feb. 20, 1989).

Following direct appeal, it was revealed in the course of state collateral review that the State had failed to disclose evidence favorable to the defense. After exhausting state remedies, Kyles sought relief on federal habeas, claiming, among other things, that the evidence withheld was material to his defense and that his conviction was thus obtained in violation of Brady. Although the United States District Court denied relief and the Fifth Circuit affirmed,6 Judge *432King dissented, writing that “[f]or the first time in my fourteen years on this court... I have serious reservations about whether the State has sentenced to death the right man.” 5 F. 3d, at 820.

Ill

The prosecution’s affirmative duty to disclose evidence favorable to a defendant can trace its origins to early 20th-century strictures against misrepresentation and is of course most prominently associated with this Court’s decision in Brady v. Maryland, 373 U. S. 83 (1963). See id., at 86 (relying on Mooney v. Holohan, 294 U. S. 103, 112 (1935), and Pyle v. Kansas, 317 U. S. 213, 215-216 (1942)). Brady held “that the suppression by the prosecution of evidence favorable to an accused upon request violates due process where the evidence is material either to guilt or to punishment, irrespective of the good faith or bad faith of the prosecution.” 373 U. S., at 87; see Moore v. Illinois, 408 U. S. 786, 794-795 *433(1972). In United States v. Agurs, 427 U. S. 97 (1976), however, it became clear that a defendant’s failure to request favorable evidence did not leave the Government free of all obligation. There, the Court distinguished three situations in which a Brady claim might arise: first, where previously undisclosed evidence revealed that the prosecution introduced trial testimony that it knew or should have known was perjured, 427 U. S., at 103-104;7 second, where the Government failed to accede to a defense request for disclosure of some specific kind of exculpatory evidence, id., at 104-107; and third, where the Government failed to volunteer exculpatory evidence never requested, or requested only in a general way. The Court found a duty on the part of the Government even in this last situation, though only when suppression of the evidence would be “of sufficient significance to result in the denial of the defendant’s right to a fair trial.” Id., at 108.

In the third prominent case on the way to current Brady law, United States v. Bagley, 473 U. S. 667 (1985), the Court disavowed any difference between exculpatory and impeachment evidence for Brady purposes, and it abandoned the distinction between the second and third Agurs circumstances, i. e., the “specific-request”'and “general- or no-request” situations. Bagley held that regardless of request, favorable evidence is material, and constitutional error results from its suppression by the government, “if there is a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been differ*434ent.” 473 U. S., at 682 (opinion of Blackmun, J.); id., at 685 (White, J., concurring in part and concurring in judgment).

Four aspects of materiality under Bagley bear emphasis. Although the constitutional duty is triggered by the potential impact of favorable but undisclosed evidence, a showing of materiality does not require demonstration by a preponderance that disclosure of the suppressed evidence would have resulted ultimately in the defendant’s acquittal (whether based on the presence of reasonable doubt or acceptance of an explanation for the crime that does not inculpate the defendant). Id., at 682 (opinion of Blackmun, J.) (adopting formulation announced in Strickland v. Washington, 466 U. S. 668, 694 (1984)); Bagley, supra, at 685 (White, J., concurring in part and concurring in judgment) (same); see 473 U. S., at 680 (opinion of Blackmun, J.) (Agurs “rejected a standard that would require the defendant to demonstrate that the evidence if disclosed probably would have resulted in acquittal”); cf. Strickland, supra, at 693 (“[W]e believe that a defendant need not show that counsel’s deficient conduct more likely than not altered the outcome in the case”); Nix v. Whiteside, 475 U. S. 157, 175 (1986) (“[A] defendant need not establish that the attorney’s deficient performance more likely than not altered the outcome in order to establish prejudice under Strickland”). Bagley’s touchstone of materiality is a “reasonable probability” of a different result, and the adjective is important. The question is not whether the defendant would more likely than not have received a different verdict with the evidence, but whether in its absence he received a fair trial, understood as a trial resulting in a verdict worthy of confidence. A “reasonable probability” of a different result is accordingly shown when the government’s evidentiary suppression “undermines confidence in the outcome of the trial.” Bagley, 473 U. S., at 678.

The second aspect of Bagley materiality bearing emphasis here is that it is not a sufficiency of evidence test. A defendant need not demonstrate that after discounting the incul*435patory evidence in light of the undisclosed evidence, there would not have been enough left to convict. The possibility of an acquittal on a criminal charge does not imply an insufficient evidentiary basis to convict. One does not show a Brady violation by demonstrating that some of the inculpatory evidence should have been excluded, but by showing that the favorable evidence could reasonably be taken to put the whole case in such a different light as to undermine confidence in the verdict.8

Third, we note that, contrary to the assumption made by the Court of Appeals, 5 F. 3d, at 818, once a reviewing court applying Bagley has found constitutional error there is no need for further harmless-error review. Assuming, arguendo, that a harmless-error enquiry were to apply, a Bagley error could not be treated as harmless, since “a reasonable probability that, had the evidence been disclosed to the defense, the result of the proceeding would have been different,” 473 U. S., at 682 (opinion of Blackmun, J.); id., at 685 (White, J., concurring in part and concurring in judgment), necessarily entails the conclusion that the suppression must have had “‘substantial and injurious effect or influence in determining the jury’s verdict,’ ” Brecht v. Abrahamson, 507 U. S. 619, 623 (1993), quoting Kotteakos v. United States, 328 U. S. 750, 776 (1946). This is amply confirmed by the development of the respective governing standards. Although *436Chapman v. California, 386 U. S. 18, 24 (1967), held that a conviction tainted by constitutional error must be set aside unless the error complained of “was harmless beyond a reasonable doubt,” we held in Brecht that the standard of harmlessness generally to be applied in habeas cases is the Kotteakos formulation (previously applicable only in reviewing nonconstitutional errors on direct appeal), Brecht, supra, at 622-623. Under Kotteakos a conviction may be set aside only if the error “had substantial and injurious effect or influence in determining the jury’s verdict.” Kotteakos, supra, at 776. Agurs, however, had previously rejected Kotteakos as the standard governing constitutional disclosure claims, reasoning that “the constitutional standard of materiality must impose a higher burden on the defendant.” Agurs, 427 U. S., at 112. Agurs thus opted for its formulation of materiality, later adopted as the test for prejudice in Strickland, only after expressly noting that this standard would recognize reversible constitutional error only when the harm to the defendant was greater than the harm sufficient for reversal under Kotteakos. In sum, once there has been Bagley error as claimed in this case, it cannot subsequently be found harmless under Brecht.9

The fourth and final aspect of Bagley materiality to be stressed here is its definition in terms of suppressed evidence considered collectively, not item by item.10 As Justice Blackmun emphasized in the portion of his opinion written for the Court, the Constitution is not violated every time the *437government fails or chooses not to disclose evidence that might prove helpful to the defense. 473 U. S., at 675, and n. 7. We have never held that the Constitution demands an open file policy (however such a policy might work out in practice), and the rule in Bagley (and, hence, in Brady) requires less of the prosecution than the ABA Standards for Criminal Justice, which call generally for prosecutorial disclosures of any evidence tending to exculpate or mitigate. See ABA Standards for Criminal Justice, Prosecution Function and Defense Function 3-3.11(a) (3d ed. 1993) (“A prosecutor should not intentionally fail to make timely disclosure to the defense, at the earliest feasible opportunity, of the existence of all evidence or information which tends to negate the guilt of the accused or mitigate the offense charged or which would tend to reduce the punishment of the accused”); ABA Model Rule of Professional Conduct 3.8(d) (1984) (“The prosecutor in a criminal case shall . . . make timely disclosure to the defense of all evidence or information known to the prosecutor that tends to negate the guilt of the accused or mitigates the offense”).

While the definition of Bagley materiality in terms of the cumulative effect of suppression must accordingly be seen as leaving the government with a degree of discretion, it must also be understood as imposing a corresponding burden. On the one side, showing that the prosecution knew of an item of favorable evidence unknown to the defense does not amount to a Brady violation, without more. But the prosecution, which alone can know what is undisclosed, must be assigned the consequent responsibility to gauge the likely net effect of all such evidence and make disclosure when the point of “reasonable probability” is reached. This in turn means that the individual prosecutor has a duty to learn of any favorable evidence known to the others acting on the government’s behalf in the case, including the police. But whether the prosecutor succeeds or fails in meeting this obligation (whether, that is, a failure to disclose is in good faith *438or bad faith, see Brady, 373 U. S., at 87), the prosecution’s responsibility for failing to disclose known, favorable evidence rising to a material level of importance is inescapable.

The State of Louisiana would prefer an even more lenient rule. It pleads that some of the favorable evidence in issue here was not disclosed even to the prosecutor until after trial, Brief for Respondent 25, 27, 30, 31, and it suggested below that it should not be held accountable under Bagley and Brady for evidence known only to police investigators and not to the prosecutor.11 To accommodate the State in this manner would, however, amount to a serious change of course from the Brady line of cases. In the State’s favor it may be said that no one doubts that police investigators sometimes fail to inform a prosecutor of all they know. But neither is there any serious doubt that “procedures and regulations can be established to carry [the prosecutor’s] burden and to insure communication of all relevant information on each case to every lawyer who deals with it.” Giglio v. United States, 405 U. S. 150, 154 (1972). Since, then, the prosecutor has the means to discharge the government’s Brady responsibility if he will, any argument for excusing a prosecutor from disclosing what he does not happen to know about boils down to a plea to substitute the police for the prosecutor, and even for the courts themselves, as the final arbiters of the government’s obligation to ensure fair trials.

Short of doing that, we were asked at oral argument to raise the threshold of materiality because the Bagley standard “makes it difficult... to know” from the “perspective [of the prosecutor at] trial... exactly what might become important later on.” Tr. of Oral Arg. 33. The State asks for “a certain amount of leeway in making a judgment call” as to the disclosure of any given piece of evidence. Ibid.

*439Uncertainty about the degree of further “leeway” that might satisfy the State’s request for a “certain amount” of it is the least of the reasons to deny the request. At bottom, what the State fails to recognize is that, with or without more leeway, the prosecution cannot be subject to any disclosure obligation without at some point having the responsibility to determine when it must act. Indeed, even if due process were thought to be violated by every failure to disclose an item of exculpatory or impeachment evidence (leaving harmless error as the government’s only fallback), the prosecutor would still be forced to make judgment calls about what would count as favorable evidence, owing to the very fact that the character of a piece of evidence as favorable will often turn on the context of the existing or potential evidentiary record. Since the prosecutor would have to exercise some judgment even if the State were subject to this most stringent disclosure obligation, it is hard to find merit in the State’s complaint over the responsibility for judgment under the existing system, which does not tax the prosecutor with error for any failure to disclose, absent a further showing of materiality. Unless, indeed, the adversary system of prosecution is to descend to a gladiatorial level unmitigated by any prosecutorial obligation for the sake of truth, the government simply cannot avoid responsibility for knowing when the suppression of evidence has come to portend such an effect on a trial’s outcome as to destroy confidence in its-result.

This means, naturally, that a prosecutor anxious about tacking too close to the wind will disclose a favorable piece of evidence. See Agurs, 427 U. S., at 108 (“[T]he prudent prosecutor will resolve doubtful questions in favor of disclosure”). This is as it should be. Such disclosure will serve to justify trust in the prosecutor as “the representative . . . of a sovereignty ... whose interest... in a criminal prosecution is not that it shall win a case, but that justice shall be done.” Berger v. United States, 295 U. S. 78, 88 (1935). *440And it will tend to preserve the criminal trial, as distinct from the prosecutor’s private deliberations, as the chosen forum for ascertaining the truth about criminal accusations. See Rose v. Clark, 478 U. S. 570, 577-578 (1986); Estes v. Texas, 381 U. S. 532, 540 (1965); United States v. Leon, 468 U. S. 897, 900-901 (1984) (recognizing general goal of establishing “procedures under which criminal defendants are ‘acquitted or convicted on the basis of all the evidence which exposes the truth’ ” (quoting Alderman v. United States, 394 U. S. 165, 175 (1969)). The prudence of the careful prosecutor should not therefore be discouraged.

There is room to debate whether the two judges in the majority in the Court of Appeals made an assessment of the cumulative effect of the evidence. Although the majority’s Brady discussion concludes with the statement that the court was not persuaded of the reasonable probability that Kyles would have obtained a favorable verdict if the jury had been “exposed to any or all of the undisclosed materials,” 5 F. 3d, at 817, the opinion also contains repeated references dismissing particular items of evidence as immaterial and so suggesting that cumulative materiality was not the touchstone. See, e.g., id., at 812 (“We do not agree that this statement made the transcript material and so mandated disclosure .... Beanie’s statement ... is itself not decisive”), 814 (“The nondisclosure of this much of the transcript was insignificant”), 815 (“Kyles has not shown on this basis that the three statements were material”), 815 (“In light of the entire record ... we cannot conclude that [police reports relating to discovery of the purse in the trash] would, in reasonable probability, have moved the jury to embrace the theory it otherwise discounted”), 816 (“We are not persuaded that these notes [relating to discovery of the gun] were material”), 816 (“[W]e are not persuaded that [the printout of the license plate numbers] would, in reasonable probability, have induced reasonable doubt where the jury did not find it. . . . the rebuttal of the photograph would have made no differ*441ence”). The result reached by the Fifth Circuit majority is compatible with a series of independent materiality evaluations, rather than the cumulative evaluation required by Bagley, as the ensuing discussion will show.

IV

In this case, disclosure of the suppressed evidence to competent counsel would have made a different result reasonably probable.

A

As the District Court put it, “the essence of the State’s case” was the testimony of eyewitnesses, who identified Kyles as Dye’s killer. 5 F. 3d, at 853 (Appendix A). Disclosure of their statements would have resulted in a markedly weaker case for the prosecution and a markedly stronger one for the defense. To begin with, the value of two of those witnesses would have been substantially reduced or destroyed.

The State rated Henry Williams as its best witness, who testified that he hád seen the struggle and the actual shooting by Kyles. The jury would have found it helpful to probe this conclusion in the light of Williams’s contemporaneous statement, in which he told the police that the assailant was “a black male, about 19 or 20 years old, about 5'4" or 5'5", 140 to 150 pounds, medium build” and that “his hair looked like it was platted.” App. 197. If cross-examined on this description, Williams would have had trouble explaining how he could have described Kyles, 6-feet tall and thin, as a man more than half a foot shorter with a medium build.12 Indeed, since Beanie was 22 years old, 5'5" tall, and 159 pounds, *442the defense would have had a compelling argument that Williams’s description pointed to Beanie but not to Kyles.13

The trial testimony of a second eyewitness, Isaac Small-wood, was equally damning to Kyles. He testified that Kyles was the assailant, and that he saw him struggle with Dye. He said he saw Kyles take a “.82, a small black gun” out of his right pocket, shoot Dye in the head, and drive off in her LTD. When the prosecutor asked him whether he actually saw Kyles shoot Dye, Smallwood answered “Yeah.” Tr. 41-48 (Dec. 6,1984).

Smallwood’s statement taken at the parking lot, however, was vastly different. Immediately after the crime, Small-*443wood claimed that he had not seen the actual murder and had not seen the assailant outside the vehicle. “I heard a lound [sic] pop,” he said. “When I looked around I saw a lady laying on the ground, and there was a red car coming toward me.” App. 189. Smallwood said that he got a look at the culprit, a black teenage male with a mustache and shoulder-length braided hair, as the victim’s red Thunderbird passed where he was standing. When a police investigator specifically asked him whether he had seen the assailant outside the car, Smallwood answered that he had not; the gunman “was already in the car and coming toward me.” Id., at 188-190.

A jury would reasonably have been troubled by the adjustments to Smallwood’s original story by the time of the second trial. The struggle and shooting, which earlier he had not seen, he was able to describe with such detailed clarity as to identify the murder weapon as a small black .32-caliber pistol, which, of course, was the type of weapon used. His description of the victim’s car had gone from a “Thunderbird” to an “LTD”; and he saw fit to say nothing about the assailant’s shoulder-length hair and moustache, details noted by no other eyewitness. These developments would have fueled a withering cross-examination, destroying confidence in Smallwood’s story and raising a substantial implication that the prosecutor had coached him to give it.14

*444Since the evolution over time of a given eyewitness’s description can be fatal to its reliability, cf. Manson v. Brathwaite, 432 U. S. 98, 114 (1977) (reliability depends in part on the accuracy of prior description); Neil v. Biggers, 409 U. S. 188, 199 (1972) (reliability of identification following impermissibly suggestive lineup depends in part on accuracy of witness’s prior description), the Smallwood and Williams identifications would have been severely undermined by use of their suppressed statements. The likely damage is best understood by taking the word of the prosecutor, who contended during closing arguments that Smallwood and Williams were the State’s two best witnesses. See Tr. of Closing Arg. 49 (Dec. 7, 1984) (After discussing Territo’s and Kersh’s testimony: “Isaac Smallwood, have you ever seen a better witness!?] • • • What’s better than that is Henry Williams. . . . Henry Williams was the closest of them all *445right here”). Nor, of course, would the harm to the State’s case on identity have been confined to their testimony alone. The fact that neither Williams nor Smallwood could have provided a consistent eyewitness description pointing to Kyles would have undercut the prosecution all the more because the remaining eyewitnesses called to testify (Territo and Kersh) had their best views of the gunman only as he fled the scene with his body partly concealed in Dye’s car. And even aside from such important details, the effective impeachment of one eyewitness can call for a new trial even though the attack does not extend directly to others, as we have said before. See Agurs, 427 U. S., at 112-113, n. 21.

B

Damage to the prosecution’s case would not have been confined to evidence of the eyewitnesses, for Beanie’s various statements would have raised opportunities to attack not only the probative value of crucial physical evidence and the circumstances in which it was found, but the thoroughness and even the good faith of the investigation, as well. By the State’s own admission, Beanie was essential to its investigation and, indeed, “made the case” against Kyles. Tr. of Closing Arg. 13 (Dec. 7,1984). Contrary to what one might hope for from such a source, however, Beanie’s statements to the police were replete with inconsistencies and would have allowed the jury to infer that Beanie was anxious to see Kyles arrested for Dye’s murder. Their disclosure would have revealed a remarkably uncritical attitude on the part of the police.

If the defense had called Beanie as an adverse witness, he could not have said anything of any significance without being trapped by his inconsistencies. A short recapitulation of some of them will make the point. In Beanie’s initial meeting with the police, and in his signed statement, he said he bought Dye’s LTD and helped Kyles retrieve his car from the Schwegmann’s lot on Friday. In his first call to the po*446lice, he said he bought the LTD on Thursday, and in his conversation with the prosecutor between trials it was again on Thursday that he said he helped Kyles retrieve Kyles’s car. Although none of the first three versions of this story mentioned Kevin Black as taking part in the retrieval of the car and transfer of groceries, after Black implicated Beanie by his testimony for the defense at the first trial, Beanie changed his story to include Black as a participant. In Beanie’s several accounts, Dye’s purse first shows up variously next to a building, in some bushes, in Kyles’s car, and at Black's house.

Even if Kyles’s lawyer had followed the more conservative course of leaving Beanie off the stand, though, the defense could have examined the police to good effect on their knowledge of Beanie’s statements and so have attacked the reliability of the investigation in failing even to consider Beanie’s possible guilt and in tolerating (if not countenancing) serious possibilities that incriminating evidence had been planted. See, e. g., Bowen v. Maynard, 799 F. 2d 593, 613 (CA10 1986) (“A common trial tactic of defense lawyers is to discredit the caliber of the investigation or the decision to charge the defendant, and we may consider such use in assessing a possible Brady violation”); Lindsey v. King, 769 F. 2d 1034, 1042 (CA5 1985) (awarding new trial of prisoner convicted in Louisiana state court because withheld Brady evidence “carried within it the potential... for the ... discrediting ... of the police methods employed in assembling the case”).15

*447By demonstrating the detectives’ knowledge of Beanie’s affirmatively self-incriminating statements, the defense could have laid the foundation for a vigorous argument that the police had been guilty of negligence. In his initial meeting with police, Beanie admitted twice that he changed the license plates on the LTD. This admission enhanced the suspiciousness of his possession of the car; the defense could have argued persuasively that he was no bona fide purchaser. And when combined with his police record, evidence of prior criminal activity near Schwegmann’s, and his status as a suspect in another murder, his devious behavior gave reason to believe that he had done more than buy a stolen car. There was further self-incrimination in Beanie’s statement that Kyles’s car was parked in the same part of the Schwegmann’s lot where Dye was killed. Beanie’s apparent awareness of the specific location of the murder could have been based, as the State contends, on television or newspaper reports, but perhaps it was not. Cf. App. 215 (Beanie saying that he knew about the murder because his brother-in-law had seen it “on T. V. and in the paper” and had told Beanie). Since the police admittedly never treated Beanie as a suspect, the defense could thus have used his statements to throw the reliability of the investigation into doubt and to sully the credibility of Detective Dillman, who testified that Beanie was never a suspect, Tr. 103-105,107 (Dec. 6,1984), and that he had “no knowledge” that Beanie had changed the license plate, id,., at 95.

The admitted failure of the police to pursue these pointers toward Beanie’s possible guilt could only have magnified the effect on the jury of explaining how the purse and the gun happened to be recovered. In Beanie’s original recorded statement, he told the police that “[Kyles’s] garbage goes out tomorrow,” and that “if he’s smart he’ll put [the purse] in [the] garbage.” App. 257. These statements, along with the internal memorandum stating that the police had “reason to believe” Dye’s personal effects and Schwegmann’s bags *448would be in the garbage, would have supported the defense’s theory that Beanie was no mere observer, but was determining the investigation’s direction and success. The potential for damage from using Beanie’s statement to undermine the ostensible integrity of the investigation is only confirmed by the prosecutor’s admission at one of Kyles’s postconviction hearings, that he did not recall a single instance before this case when police had searched and seized garbage on the street in front of a residence, Tr. of Hearing on Post-Conviction Relief 113 (Feb. 20, 1989), and by Detective John Miller’s admission at the same hearing that he thought at the time that it “was a possibility” that Beanie had planted the incriminating evidence in the garbage, Tr. of Hearing on Post-Conviction Relief 51 (Feb. 24, 1989). If a police officer thought so, a juror would have, too.16

To the same effect would have been an enquiry based on Beanie’s apparently revealing remark to police that “if you can set [Kyles] up good, you can get that same gun.”17 App. 228-229. While the jury might have understood that Beanie meant simply that if the police investigated Kyles, they would probably find the murder weapon, the jury could also have taken Beanie to have been making the more sinister *449suggestion that the police “set up” Kyles, and the defense could have argued that the police accepted the invitation. The prosecutor’s notes of his interview with Beanie would have shown that police officers were asking Beanie the whereabouts of the gun all day Sunday, the very day when he was twice at Kyles’s apartment and was allegedly seen by Johnny Burns lurking near the stove, where the gun was later found.18 Beanie’s same statement, indeed, could have been used to cap an attack on the integrity of the investigation and on the reliability of Detective Dillman, who testified on cross-examination that he did not know if Beanie had been at Kyles’s apartment on Sunday. Tr. 93, 101 (Dec. 6,1984).19

*450C

Next to be considered is the prosecution’s list of the cars in the Schwegmann’s parking lot at mid-evening after the murder. While its suppression does not rank with the failure to disclose the other evidence discussed here, it would have had some value as exculpation and impeachment, and it counts accordingly in determining whether Bagley’s standard of materiality is satisfied. On the police’s assumption, argued to the jury, that the killer drove to the lot and left his car there during the heat of the investigation, the list without Kyles’s registration would obviously have helped Kyles and would have had some value in countering an argument by the prosecution that a grainy enlargement of a photograph of the crime scene showed Kyles’s car in the background. The list would also have shown that the police either knew that it was inconsistent with their informant’s second and third statements (in which Beanie described retrieving Kyles’s car after the time the list was compiled) or never even bothered to check the informant’s story against known fact. Either way, the defense would have had further support for arguing that the police were irresponsible in relying on Beanie to tip them off to the location of evidence damaging to Kyles.

The State argues that the list was neither impeachment nor exculpatory evidence because Kyles could have moved his car before the list was created and because the list does *451not purport to be a comprehensive listing of all the cars in the Schwegmann’s lot. Such argument, however, confuses the weight of the evidence with its favorable tendency, and even if accepted would work against the State, not for it. If the police had testified that the list was incomplete, they would simply have underscored the unreliability of the investigation and complemented the defense’s attack on the failure to treat Beanie as a suspect and his statements with a presumption of fallibility. But however the evidence would have been used, it would have had some weight and its tendency would have been favorable to Kyles.

D

In assessing the significance of the evidence withheld, one must of course bear in mind that not every item of the State’s case would have been directly undercut if the Brady evidence had been disclosed. It is significant, however, that the physical evidence remaining unscathed would, by the State’s own admission, hardly have amounted to overwhelming proof that Kyles was the murderer. See Tr. of Oral Arg. 56 (“The heart of the State’s case was eye-witness identification”); see also Tr. of Hearing on Post-Conviction Relief 117 (Feb. 20, 1989) (testimony of chief prosecutor Strider) (“The crux of the case was the four eye-witnesses”). Ammunition and a holster were found in Kyles’s apartment, but if the jury had suspected the gun had been planted the significance of these items might have been left in doubt. The fact that pet food was found in Kyles’s apartment was consistent with the testimony of several defense witnesses that Kyles owned a dog and that his children fed stray cats. The brands of pet food found were only two of the brands that Dye typically bought, and these two were common, whereas the one specialty brand that was found in Dye’s apartment after her murder, Tr. 180 (Dec. 7,1984), was not found in Kyles’s apartment, id., at 188. Although Kyles was wrong in describing the cat food as being on sale the day he said he bought it, he *452was right in describing the way it was priced at Schwegmann’s market, where he commonly shopped.20

Similarly undispositive is the small Schwegmann’s receipt on the front passenger floorboard of the LTD, the only physical evidence that bore a fingerprint identified as Kyles’s, Kyles explained that Beanie had driven him to Schwegmann’s on Friday to buy cigarettes and transmission fluid, and he theorized that the slip must have fallen out of the bag when he removed the cigarettes. This explanation is consistent with the location of the slip when found and with its small size. The State cannot very well argue that the fingerprint ties Kyles to the killing without also explaining how the 2-inch-long register slip could have been the receipt for a week’s worth of groceries, which Dye had gone to Schwegmann’s to purchase. Id., at 181-182.21

*453The inconclusiveness of the physical evidence does not, to be sure, prove Kyles’s innocence, and the jury might have found the eyewitness testimony of Territo and Kersh sufficient to convict, even though less damning to Kyles than that of Smallwood and Williams.22 But the question is not whether the State would have had a case to go to the jury if it had disclosed the favorable evidence, but whether we can be confident that the jury’s verdict would have been the same. Confidence that it would have been cannot survive a recap of the suppressed evidence and its significance for the prosecution. The jury would have been entitled to find

(a) that the investigation was limited by the police’s uncritical readiness to accept the story and suggestions of an informant whose accounts were inconsistent to the point, for example, of including four different versions of the discovery of the victim’s purse, and whose own behavior was enough to raise suspicions of guilt;

(b) that the lead police detective who testified was either less than wholly candid or less than fully informed;

(c) that the informant’s behavior raised suspicions that he had planted both the murder weapon and the victim’s purse in the places they were found;

(d) that one of the four eyewitnesses crucial to the State’s case had given a description that did not match the defendant and better described the informant;

(e) that another eyewitness had been coached, since he had first stated that he had not seen the killer outside the getaway car, or the killing itself, whereas at trial he *454claimed to have seen the shooting, described the murder weapon exactly, and omitted portions of his initial description that would have been troublesome for the case; (f) that there was no consistency to eyewitness descriptions of the killer’s height, build, age, facial hair, or hair length.

Since all of these possible findings were precluded by the prosecution’s failure to disclose the evidence that would have supported them, “fairness” cannot be stretched to the point of calling this a fair trial. Perhaps, confidence that the verdict would have been the same could survive the evidence impeaching even two eyewitnesses if the discoveries of gun and purse were above suspicion. Perhaps those suspicious circumstances would not defeat confidence in the verdict if the eyewitnesses had generally agreed on a description and were free of impeachment. But confidence that the verdict would have been unaffected cannot survive when suppressed evidence would have entitled a jury to find that the eyewitnesses were not consistent in describing the killer, that two out of the four eyewitnesses testifying were unreliable, that the most damning physical evidence was subject to suspicion, that the investigation that produced it was insufficiently probing, and that the principal police witness was insufficiently informed or candid. This is not the “massive” case envisioned by the dissent, post, at 475; it is a significantly weaker case than the one heard by the first jury, which could not even reach a verdict.

The judgment of the Court of Appeals is reversed, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.

concurring.

As the Court has explained, this case presents an important legal issue. See ante, at 440-441. Because Justice *455Scalia so emphatically disagrees, I add this brief response to his criticism of the Court’s decision to grant certiorari.

Proper management of our certiorari docket, as Justice Scalia notes, see post, at 456-460, precludes us from hearing argument on the merits of even a “substantial percentage” of the capital cases that confront us. Compare Coleman v. Balkcom, 451 U. S. 949 (1981) (Stevens, J., concurring in denial of certiorari), with id., at 956 (Rehnquist, J., dissenting). Even aside from its legal importance, however, this case merits “favored treatment,” cf. post, at 457, for at least three reasons. First, the fact that the jury was unable to reach a verdict at the conclusion of the first trial provides strong reason to believe the significant errors that occurred at the second trial were prejudicial. Second, cases in which the record reveals so many instances of the state’s failure to disclose exculpatory evidence are extremely rare. Even if I shared Justice Scalia’s appraisal of the evidence in this case — which I do not — I would still believe we should independently review the record to ensure that the prosecution’s blatant and repeated violations of a well-settled constitutional obligation did not deprive petitioner of a fair trial. Third, despite my high regard for the diligence and craftsmanship of the author of the majority opinion in the Court of Appeals, my independent review of the case left me with the same degree of doubt about petitioner’s guilt expressed by the dissenting judge in that court.

Our duty to administer justice occasionally requires busy judges to engage in a detailed review of the particular facts of a case, even though our labors may not provide posterity with a newly minted rule of law. The current popularity of capital punishment makes this “generalizable principle,” post, at 460, especially important. Cf. Harris v. Alabama, 513 U. S. 504, 519-520, and n. 5 (1995) (Stevens, J., dissenting). I wish such review were unnecessary, but I cannot agree that our position in the judicial hierarchy makes it inappropriate. Sometimes the performance of an unpleasant *456duty conveys a message more significant than even the most penetrating legal analysis.