delivered the opinion of the Court.

Under Cage v. Louisiana, 498 U. S. 39 (1990) (per curiam), a jury instruction is unconstitutional if there is a reasonable likelihood that the jury understood the instruction to allow conviction without proof beyond a reasonable doubt.1 In *659this case, we must decide whether this rule was “made retroactive to cases on collateral review by the Supreme Court.” 28 U. S. C. § 2244(b)(2)(A) (1994 ed., Supp. V). We hold that it was not.

I

During a fight with his estranged girlfriend in March 1975, petitioner Melvin Tyler shot and killed their 20-day-old daughter. A jury found Tyler guilty of second-degree murder, and his conviction was affirmed on appeal. After sentencing, Tyler assiduously sought postconviction relief. By 1986, he had filed five state petitions, all of which were denied. See State ex rel. Tyler v. Blackburn, 494 So. 2d 1171 (La. 1986); State v. Tyler, 446 So. 2d 1226 (La. 1984); State ex rel. Tyler v. State, 437 So. 2d 1142 (La. 1983); State v. Tyler, 430 So. 2d 92 (La. 1983); State ex rel. Tyler v. Maggio, 428 So. 2d 483 (La. 1982). He next filed a federal habeas petition, which was unsuccessful as well. Tyler v. Butler, No. 88cv4929 (ED La.), aff’d, Tyler v. Whitley, 920 F. 2d 929 (CA5 1990). After this Court’s decision in Cage, Tyler continued his efforts. Because the jury instruction defining reasonable doubt at Tyler’s trial was substantively identical to the instruction condemned in Cage, Tyler filed a sixth state postconviction petition, this time raising a Cage claim. The State District Court denied relief, and the Louisiana Supreme Court affirmed. State ex rel. Tyler v. Cain, 684 So. 2d 950 (1996).

In early 1997, Tyler returned to federal court. Seeking to pursue his Cage claim, Tyler moved the United States *660Court of Appeals for the Fifth Circuit for permission to file a second habeas corpus application, as required by the Anti-terrorism and Effective Death Penalty Act of 1996 (AEDPA), 110 Stat. 1214.2 The Court of Appeals recognized that it could not grant the motion unless Tyler made “a prima facie showing,” § 2244(b)(3)(C), that his “claim relies on a new rule of constitutional law, made retroactive to eases on collateral review by the Supreme Court, that was previously unavailable,” § 2244(b)(2)(A). Finding that Tyler had made the requisite prima facie showing, the Court of Appeals granted the motion, thereby allowing Tyler to file a habeas petition in District Court.

The District Court proceeded to the merits of Tyler’s claim and held that, although Cage should apply retroactively, App. 5-7 (citing Humphrey v. Cain, 138 F. 3d 552 (CA5 1998) (en banc)), Tyler was not entitled to collateral relief. Under AEDPA, a state prisoner can prevail only if the state court’s decision “was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States.” § 2254(d)(1). Concluding that Tyler could not overcome this barrier, the District Court denied his petition.

The Court of Appeals affirmed. Judgt. order reported at 218 F. 3d 744 (CA5 2000). It stated, however, that the District Court erred by failing first to determine whether Tyler “satisfied AEDPA’s successive habeas standard.” App. 15. AEDPA requires a district court to dismiss a claim in a second or successive application unless, as relevant here, the applicant “shows” that the “claim relies on a new rule of constitutional law, made retroactive to cases on collateral review by the Supreme Court, that was previously unavailable,”*6613 §2244(b)(2)(A) (emphasis added); §2244(b)(4). Relying on Circuit precedent, see Brown v. Leming, 171 F. 3d 1031 (CA5 1999); In re Smith, 142 F. 3d 832 (CA5 1998), the Court of Appeals concluded that Tyler did not meet this standard because he “could not show that any Supreme Court decision renders the Cage decision retroactively applicable to cases on collateral review.” App. 15.

The Courts of Appeals are divided on the question whether Cage was “made retroactive to cases on collateral review by the Supreme Court,” as required by 28 U. S. C. § 2244(b)(2)(A). Compare Rodriguez v. Superintendent, 139 F. 3d 270 (CA1 1998) (holding that Cage has not been made retroactive by the Supreme Court); Brown, supra (same); In re Hill, 113 F. 3d 181 (CA11 1997) (same), with West v. Vaughn, 204 F. 3d 53 (CA3 2000) (holding that Cage has been made retroactive to cases on collateral review). To resolve this conflict, we granted certiorari. 531 U. S. 1051 (2000).

h — <

AEDPA greatly restricts the power of federal courts to award relief to state prisoners who file second or successive habeas corpus applications. If the prisoner asserts a claim that he has already presented in a previous federal ha-beas petition, the claim must be dismissed in all cases. § 2244(b)(1). And if the prisoner asserts a claim that was not presented in a previous petition, the claim must be dismissed unless it falls within one of two narrow exceptions. One of these exceptions is for claims predicated on newly *662discovered facts that call into question the accuracy of a guilty verdict. § 2244(b)(2)(B). The other is for certain claims relying on new rules of constitutional law. § 2244(b)(2)(A).

It is the latter exception that concerns us today. Specifically, § 2244(b)(2)(A) covers claims that “rel[y] on a new rule of constitutional law, made retroactive to cases on collateral review by the Supreme Court, that was previously unavailable.” This provision establishes three prerequisites to obtaining relief in a second or successive petition: First, the rule on which the claim relies must be a “new rule” of constitutional law; second, the rule must have been “made retroactive to cases on collateral review by the Supreme Court”; and third, the claim must have been “previously unavailable.” In this case, the parties ask us to interpret only the second requirement; respondent does not dispute that Cage created a “new rule” that was “previously unavailable.” Based on the plain meaning of the text read as a whole, we conclude that “made” means “held” and, thus, the requirement is satisfied only if this Court has held that the new rule is retroactively applicable to cases on collateral review.

A

As commonly defined, “made” has several alternative meanings, none of which is entirely free from ambiguity. See, e. g., Webster’s Ninth New Collegiate Dictionary 718-719 (1991) (defining “to make” as “to cause to happen,” “to cause to exist, occur or appear,” “to lay out and construct,” and “to cause to act in a certain way”). Out of context, it may thus be unclear which meaning should apply in § 2244(b)(2)(A), and how the term should be understood. We do not, however, construe the meaning of statutory terms in a vacuum. Rather, we interpret the words “in their context and with a view to their place in the overall statutory scheme.” Davis v. Michigan Dept. of Treasury, 489 U. S. 803, 809 (1989). In § 2244(b)(2)(A), the word “made” falls within a clause that reads as follows: “[A] new rule of consti*663tutional law, made retroactive to cases on collateral review by the Supreme Court.” (Emphasis added.) Quite significantly, under this provision, the Supreme Court is the only entity that can “ma[k]e” a new rule retroactive. The new rule becomes retroactive, not by the decisions of the lower court or by the combined action of the Supreme Court and the lower courts, but simply by the action of the Supreme Court.

The only way the Supreme Court can, by itself, “lay out and construct” a rule’s retroactive effect, or “cause” that effect “to exist, occur, or appear,” is through a holding. The Supreme Court does not “ma[k]e” a rule retroactive when it merely , establishes principles of retroactivity and leaves the application of those principles to lower courts. In such an event, any legal conclusion that is derived from the principles is developed by the lower court (or perhaps by a combination of courts), not by the Supreme Court.4 We thus conclude that a new rule is not “made retroactive to cases on collateral review” unless the Supreme Court holds it to be retroactive.5

*664To be sure, the statute uses the word “made,” not “held.” But we have already stated, in a decision interpreting another provision of AEDPA, that Congress need not use the word “held” to require as much. In Williams v. Taylor, 529 U. S. 362 (2000), we concluded that the phrase “clearly established Federal law, as determined by the Supreme Court of the United States,” § 2254(d)(1) (emphasis added), “refers to the holdings, as opposed to the dicta, of this Court’s decisions,” id., at 412. The provision did not use the word “held,” but the effect was the same. Congress, needless to say, is permitted to use synonyms in a statute. And just as “determined” and “held” are synonyms in the context of § 2254(d)(1), “made” and “held” are synonyms in the context of § 2244(b)(2)(A).

We further note that our interpretation is necessary for the proper implementation of the collateral review structure created by AEDPA. Under the statute, before a state prisoner may file a second or successive habeas application, he “shall move in the appropriate court of appeals for an order authorizing the district court to consider the application.” § 2244(b)(3)(A). The court of appeals must make a decision on the application within 30 days. § 2244(b)(3)(D). In this limited time, the court of appeals must determine whether the application “makes a prima facie showing that [it] satisfies the [second habeas standard].” § 2244(b)(3)(C). It is unlikely that a court of appeals could make such a determination in the allotted time if it had to do more than simply rely on Supreme Court holdings on retroactivity. The stringent time limit thus suggests that the courts of appeals do not have to engage in the difficult legal analysis that can be required to determine questions of retroactivity in the first instance.

B

Because “made” means “held” for purposes of § 2244(b)(2)(A), it is clear that the Cage rule has not been “made retroactive to cases on collateral review by the Supreme Court.” Cage itself does not hold that it is retroac*665tive. The only holding in Cage is that the particular jury instruction violated the Due Process Clause.

Tyler argues, however, that a subsequent case, Sullivan v. Louisiana, 508 U. S. 275 (1993), made the Cage rule retroactive. But Sullivan held only that a Cage error is structural — i. e., it is not amenable to harmless-error analysis and “will always invalidate the conviction.” 508 U. S., at 279. Conceding that the holding in Sullivan does not render Cage retroactive to cases on collateral review, Tyler contends that the reasoning in Sullivan makes clear that retroactive application is warranted by the principles of Teague v. Lane, 489 U. S. 288 (1989). Under Teague, a new rule can be retroactive to cases on collateral review if, and only if, it falls within one of two narrow exceptions to the general rule of nonretro-activity. Id., at 311-313 (plurality opinion). See also O’Dell v. Netherland, 521 U. S. 151, 156-157 (1997). The exception relevant here is for “watershed rules of criminal procedure implicating the fundamental fairness and accuracy of the criminal proceeding.” Graham v. Collins, 506 U. S. 461, 478 (1993). To fall within this exception, a new rule must meet two requirements: Infringement of the rule must “seriously diminish the likelihood of obtaining an accurate conviction,” and the rule must “' “alter our understanding of the bedrock procedural elements” ’ essential to the fairness of a proceeding.” Sawyer v. Smith, 497 U. S. 227, 242 (1990) (quoting Teague, supra, at 311 (plurality opinion), in turn quoting Mackey v. United States, 401 U. S. 667, 693 (1971) (Harlan, J., concurring in judgments in part and dissenting in part)).

According to Tyler, the reasoning of Sullivan demonstrates that the Cage rule satisfies both prongs of this Teague exception. First, Tyler notes, Sullivan repeatedly emphasized that a Cage error fundamentally undermines the reliability of a trial’s outcome. And second, Tyler contends, the central point of Sullivan is that a Cage error deprives a defendant of a bedrock element of procedural fairness: the right to have the jury make the determination of guilt beyond a reasonable doubt. Tyler’s arguments fail to persuade, how*666ever. The most he can claim is that, based on the principles outlined in Teague, this Court should make Cage retroactive to cases on collateral review. What is clear, however, is that we have not “made” Cage retroactive to eases on collateral review.6

Justice Breyer observes that this Court can make a rule retroactive over the course of two cases. See post, at 672-673 (dissenting opinion). We do not disagree that, with the right combination of holdings, the Court could do this. But even so, the Court has not made Cage retroactive. Multiple cases can render a new rule retroactive only if the holdings in those cases necessarily dictate retroactivity of the new rule. The only holding in Sullivan is that a Cage error is structural error. There is no second case that held that all structural-error rules apply retroactively or that all structural-error rules fit within the second Teague exception. The standard for determining whether an error is structural, see generally Arizona v. Fulminante, 499 U. S. 279 (1991), is not coextensive with the second Teague exception,7 and a *667holding that a particular error is structural does not logically dictate the conclusion that the second Teague exception has been met.

Ill

Finally, Tyler suggests that, if Cage has not been made retroactive to cases on collateral review, we should make it retroactive today. We disagree. Because Tyler’s habeas application was his second, the District Court was required to dismiss it unless Tyler showed that this Court already had made Cage retroactive. § 2244(b)(4) (“A district court shall dismiss any claim presented in a second or successive application that the court of appeals has authorized to be filed unless the applicant shows that the claim satisfies the requirements of this section”); § 2244(b)(2)(A) (“A claim presented in a second or successive habeas corpus application under section 2254 that was not presented in a prior application shall be dismissed unless . . . the applicant shows that *668the claim relies on a new rule of constitutional law, made retroactive to eases on collateral review by the Supreme Court, that was previously unavailable”). We cannot decide today whether Cage is retroactive to eases on collateral review, because that decision would not help Tyler in this case. Any statement on Cage’s retroactivity would be dictum, so we decline to comment further on the issue.

* * *

The judgment of the Court of Appeals is affirmed.

It is so ordered.

concurring.

I join the Court’s opinion and write separately to explain more fully the circumstances in which a new rule is “made retroactive to cases on collateral review by the Supreme Court.” 28 U S. C. § 2244(b)(2)(A) (1994 ed., Supp. V).

It is only through the holdings of this Court, as opposed to this Court’s dicta and as opposed to the decisions of any other court, that a new rule is “made retroactive ... by the Supreme Court” within the meaning of § 2244(b)(2)(A). See ante, at 663; cf. Williams v. Taylor, 529 U. S. 362, 412 (2000). The clearest instance, of course, in which we can be said to have “made” a new rule retroactive is wheré we expressly have held the new rule to be retroactive in a. case on collateral review and applied the rule to that case. But, as the Court recognizes, a single case that expressly holds a rule to be retroactive is not a sine qua non for the satisfaction of this statutory provision. Ante, at 666. This Court instead may “ma[k]e” a new rule retroactive through multiple holdings that logically dictate the retroactivity of the new rule. Ibid. To apply the syllogistic relationship described by Justice Breyer, post, at 672-673 (dissenting opinion), if we hold in Case One that a particular type of rule applies retroactively to cases on collateral review and hold in Case Two *669that a given rule is of that particular type, then it necessarily follows that the given rule applies retroactively to cases on collateral review. In such circumstances, we can be said to have “made” the given rule retroactive to cases on collateral review.

The relationship between the conclusion that a new rule is retroactive and the holdings that “ma[k]e” this rule retroactive, however, must be strictly logical — i e., the holdings must dictate the conclusion and not merely provide principles from which one may conclude that the rule applies retroactively. As the Court observes, “[t]he Supreme Court does not ‘ma[k]e’ a rule retroactive when it merely establishes principles of retroactivity and leaves the application of those principles to lower courts.” Ante, at 663. The Court instead can be said to have “made” a rule retroactive within the meaning of § 2244(b)(2)(A) only where the Court’s holdings logically permit no other conclusion than that the rule is retroactive.

It is relatively easy to demonstrate the required logical relationship with respect to the first exception articulated in Teague v. Lane, 489 U. S. 288 (1989). Under this exception, “a new rule should be applied retroactively if it places ‘certain kinds of primary, private individual conduct beyond the power of the criminal law-making authority to proscribe.’” Id., at 307 (plurality opinion) (quoting Mackey v. United States, 401 U. S. 667, 692 (1971) (Harlan, J., concurring in judgments in part and dissenting in part)). When the Court holds as a new rule in a subsequent case that a particular species of primary, private individual conduct is beyond the power of the criminal lawmaking authority to proscribe, it necessarily follows that this Court has “made” that new rule retroactive to eases on collateral review. The Court has done so through its holdings alone, without resort to dicta and without any application of principles by lower courts.

The matter is less straightforward with respect to the second Teague exception, which is reserved for “watershed *670rules of criminal procedure,” 489 U. S., at 311 (plurality opinion). A case announcing a new rule could conceivably hold that infringement of the rule “seriously diminishes] the likelihood of obtaining an accurate conviction,” id., at 815, and that the rule “ ‘alter[s] our understanding of the bedrock procedural elements essential to the fairness of a proceeding/ ” id., at 311 (plurality opinion) (quoting Mackey, supra, at 693 (Harlan, J., concurring in judgments in part and dissenting in part)); see also Sawyer v. Smith, 497 U. S. 227, 242 (1990), without holding in so many words that the rule “applies retroactively” and without actually applying that rule retroactively to a case on collateral review. The “precise contours” of this Teague exception, of course, “may be difficult to discern,” Saffle v. Parks, 494 U. S. 484,495 (1990), and the judgment involved in our “ma[king]” a new rule retroactive under this exception is likely to be more subjective and self-conscious than is the case with Teague’s first exception. But the relevant inquiry is not whether the new rule comes within the Teague exception at all, but the more narrow and manageable inquiry of whether this Court’s holdings, by strict logical necessity, “ma[k]e” the new rule retroactive within the meaning of § 2244(b)(2)(A). While such logical necessity does not obtain in this particular case, ante, at 665-667, this Court could “ma[k]e” a new rule retroactive under Teague’s second exception in this manner.

with Ayhom Justice Stevens, Justice Souter, and Justice Ginsburg join, dissenting.

In Cage v. Louisiana, 498 U. S. 39 (1990) (per curiam), this Court held that a certain jury instruction violated the Constitution because it inaccurately defined “reasonable doubt,” thereby permitting a jury to convict “based on a degree of proof below that required by the Due Process Clause.” Id., at 41. Here we must decide whether this Court has “made” Cage “retroactive to cases on collateral *671review.” 28 U. S. C. § 2244(b)(2)(A) (1994 ed., Supp. V). I believe that it has.

The Court made Cage retroactive in two cases taken together. Case One is Teague v. Lane, 489 U. S. 288 (1989). That case, as the majority says, held (among other things) that a new rule is applicable retroactively to cases on collateral review if (1) infringement of the new rule will “seriously diminish the likelihood of obtaining an accurate conviction,” id., at 315 (plurality opinion), and (2) the new rule “ ‘alter[s] our understanding of the bedrock procedural elements that must be found to vitiate the fairness of a particular conviction,’” id., at 311 (plurality opinion) (quoting Mackey v. United States, 401 U. S. 667, 693 (1971) (Harlan, J., concurring in judgments in part and dissenting in part)) (emphasis deleted).

Case Two is Sullivan v. Louisiana, 508 U. S. 275 (1993). This Court decided Sullivan after several lower courts had held that Cage’s rule did not fall within the Teague “watershed” exception I have just mentioned. See, e. g., Adams v. Aiken, 965 F. 2d 1306, 1312 (CA4 1992), vacated, 511 U. S. 1001 (1994); Skelton v. Whitley, 950 F. 2d 1037, 1045 (CA5), cert, denied, 506 U. S. 833 (1992). The question in Sullivan was whether a violation of the Cage rule could ever count as harmless error. The Court answered that question in the negative. In so concluding, the Court reasoned that an instruction that violated Cage by misdescribing the concept of reasonable doubt “vitiates all the jury’s findings,” and deprives a criminal defendant of a “basic protection... without which a criminal trial cannot reliably serve its function.” Sullivan, supra, at 281 (emphasis in original; internal quotation marks omitted). It renders the situation as if “there has been no jury verdict within the meaning of the Sixth Amendment.” 508 U. S., at 280.

, To reason as the Court reasoned in Sullivan is to hold (in Teague’s language) (1) that infringement of the Cage rule “seriously diminishes] the likelihood of obtaining an accurate *672conviction,” Teague, supra, at 315 (plurality opinion), and (2) that Cage “alter[s] our understanding of the bedrock procedural elements” that are essential to the fairness of a criminal trial, 489 U. S., at 311 (plurality opinion) (internal quotation marks omitted; emphasis deleted). That is because an instruction that makes “all the jury’s findings” untrustworthy, Sullivan, supra, at 281, must “diminish the likelihood of obtaining an accurate conviction,” Teague, supra, at 315 (plurality opinion). It is because a deprivation of a “basic protection” needed for a trial to “serve its function,” Sullivan, supra, at 281 (internal quotation marks omitted), is a deprivation of a “bedrock procedural element,” Teague, supra, at 311 (plurality opinion) (internal quotation marks omitted). And it is because Cage significantly “alter[ed]” pre-existing law. 489 U. S., at 311. That is what every Court of Appeals to have considered the matter has concluded. See Tillman v. Cook, 215 F. 3d 1116, 1122 (CA10), cert. denied, 531 U. S. 1055 (2000); West v. Vaughn, 204 F. 3d 53, 61, and n. 9 (CA3 2000); Gaines v. Kelly, 202 F. 3d 598, 604-605 (CA2 2000); Humphrey v. Cain, 138 F. 3d 552, 553 (CA5) (en banc), cert. denied, 525 U. S. 935 (1998); Adams v. Aiken, 41 F. 3d 175, 178-179 (CA4 1994), cert. denied, 515 U. S. 1124 (1995); Nutter v. White, 39 F. 3d 1154, 1158 (CA11 1994). But cf. In re Smith, 142 F. 3d 832, 835-836 (CA5 1998) (concluding that explicit Supreme Court statement is necessary to make Cage retroactive for second or successive habeas purposes); Rodriguez v. Superintendent, Bay State Correctional Ctr., 139 F. 3d 270, 275-276 (CA1 1998) (same); In re Hill, 113 F. 3d 181, 184 (CA11 1997) (same). And I do not see how the majority can deny that this is so.

Consequently, Sullivan, in holding that a Cage violation can never be harmless because it leaves the defendant with no jury verdict known to the Sixth Amendment, also holds that Cage falls within Teague's “watershed” exception. The matter is one of logic. If Case One holds that all men are *673mortal and Case Two holds that Socrates is a man, we do not need Case Three to hold that Socrates is mortal. It is also a matter of law. If Case One holds that a party’s expectation measures damages for breach of contract and Case Two holds that Circumstances X, Y, and Z create a binding contract, we do not need Case Three to hold that in those same circumstances expectation damages are awarded for breach. Ordinarily, in law, to hold that a set of circumstances falls within a particular legal category is simultaneously to hold that, other things being equal, the normal legal characteristics of members of that category apply to those circumstances.

The majority says that Sullivan’s only “holding” is that Cage error is structural, and that this “holding” does not dictate the “watershed” nature of the Cage rule. See ante, at 665-666. ' But the majority fails to identify a meaningful difference between the definition of a watershed rule under Teague and the standard that we have articulated in the handful of instances in which we have held errors structural, namely, that structural errors deprive a defendant of a “‘basic protectio[n]’” without which a “‘trial cannot reliably serve its function as a vehicle for determination of guilt or innocence’” to the point where “‘no criminal punishment may be regarded as fundamentally fair.’ ” Arizona v. Fulminante, 499 U. S. 279, 310 (1991) (quoting Rose v. Clark, 478 U. S. 570, 577-578 (1986)); see also Neder v. United States, 527 U. S. 1, 8 (1999) (identifying the six kinds of error, including Cage error, that have been held structural). In principle Teague also adds an element that “structural error” alone need not encompass, namely, the requirement that a violation of the rule must undermine accuracy. But that additional accuracy requirement poses no problem here, for our language in Sullivan could not have made clearer that Cage error seriously undermines the accuracy and reliability of a guilty verdict.

Of course, as the majority points out, identifying an error as structural need not “alter our understanding of th[e] fun*674damental procedural elements” that are essential to a fair trial. See ante, at 666, n. 7. But this “altering” requirement is not a problem here. No one denies that Cage’s rule ; was a new one. “Whether a trial court’s unconstitutional misdescription of the burden of proof in a criminal case violates the Due Process Clause was certainly an open question before Cage.” Adams, 41 F. 3d, at 178; see also Gaines, supra, at 606-607 (noting that Cage led to reversals of numerous convictions that had been based on similar reasonable-doubt instruction); State v. Humphrey, 544 So. 2d 1188, 1192 (La. App.) (citing multiple decisions by Louisiana Supreme Court which had upheld reasonable-doubt instructions like that invalidated in Cage), cert. denied, 550 So. 2d 627 (1989). And our holding that such a misdescription of the burden of proof means that “there has been no jury verdict within the meaning of the Sixth Amendment,” Sullivan, 508 U. S., at 280, certainly altered the understanding of the significance of such an error.

Insofar as the majority means to suggest that a rule may be sufficiently “new” that it does not apply retroactively but not “new enough” to qualify for the watershed exception, I note only that the cases establishing this exception suggest no such requirement. Rather than focus on the “degree of newness” of a new rule, these decisions. emphasize that watershed rules are those that form part of the fundamental requirements of due process. See Teague, 489 U. S., at 311-312 (plurality opinion); Mackey, 401 U. S., at 693-694 (Harlan, J., concurring in judgments in part and dissenting in part); cf. O’Dell v. Netherland, 521 U. S. 151, 167 (1997) (holding that “narrow right of rebuttal” established by Simmons v. South Carolina, 512 U. S. 154 (1994), “has hardly alter[ed] our understanding of the bedrock procedural elements essential to the fairness of a proceeding” (internal quotation marks omitted; emphasis in original)); Caspari v. Bohlen, 510 U. S. 383, 396 (1994) (holding that application of double jeopardy bar to successive noncapital sentencing would not be unfair and would enhance rather than hinder *675accuracy); Sawyer v. Smith, 497 U. S. 227, 242-244 (1990) (holding that rule which “provided] an additional measure of protection” to existing prohibition on prosecutorial remarks that render a proceeding “fundamentally unfair” was not “an ‘absolute prerequisite to fundamental fairness’ ” that would fall within the second Teague exception) (quoting Teague, supra, at 314 (plurality opinion)).

Nor does the majority explain why the reasoning that was necessary to our holding in Sullivan (and is therefore binding upon all courts) lacks enough legal force to “make” the Cage rule retroactive. Cf. Seminole Tribe of Fla. v. Florida, 517 U. S. 44, 67 (1996) (“We adhere... not to mere obiter dicta, but rather to the well-established rationale upon which the Court based the results of its earlier decisions. When an opinion issues for the Court, it is not only the result but also those portions of the opinion necessary to that result by which we are bound”); Burnham v. Superior Court of Cal, County of Marin, 495 U. S. 604, 613, n. 2 (1990) (plurality opinion) (exclusive basis for judgment is not dicta). In any event, technical issues about what constitutes a “holding” are beside the point. The statutory provision before us does not use the words “holding” or “held.” But cf. ante, at 664 (majority opinion) (stating without explanation that “made” means “held”). It uses the word “made.” It refers to instances in which the Supreme Court has “mode” a rule of law “retroactive to cases on collateral review.” 28 U. S. C. § 2244(b)(2)(A) (1994 ed., Supp. V) (emphasis added). And that is just what the Supreme Court, through Teague and Sullivan, has done with respect to the rule of Cage.

I agree with Justice O’Connor — as does a majority of the Court — when (in describing a different Teague exception) she says that “[w]hen the Court holds as a new rule in a subsequent case that a particular species of primary, private individual conduct is beyond the power, of the criminal lawmaking authority to proscribe, it necessarily follows that this Court has ‘made’ that new rule retroactive to cases on collateral review.” Ante, at 669 (concurring opinion). But I do *676not understand why a decision by this Court which makes it apparent that a rule is retroactive under Teague’s second exception will necessarily be “more subjective and self-conscious.” Ante, at 670 (concurring opinion). Of course, it will sometimes be difficult to decide whether an earlier Supreme Court case has satisfied the watershed rule’s requirements. But that is not so here. In Sullivan, this Court used language that unmistakably stated that a defective reasonable-doubt instruction undermines the accuracy of a trial and deprives the defendant of a bedrock element that is essential to the fairness of a criminal proceeding. That is sufficient to make Teague’s watershed exception applicable.

I would add two further points. First, nothing in the statute’s purpose favors, let alone requires, the majority’s conclusion. That purpose, as far as I can surmise, is to bar successive petitions when lower courts, but not the Supreme Court, have held a rule not to be “new” under Teague because dictated by their own precedent, cf. Dyer v. Calderon, 151 F. 3d 970, 993-995 (CA9) (en banc) (O’Scannlain, J., dissenting) (rejecting proposition that lower court decisions can establish rule for Teague purposes), cert. denied, 525 U. S. 1033 (1998); Clemmons v. Delo, 124 F. 3d 944, 955, n. 11 (CA8 1997) (assuming, without deciding, that only Supreme Court precedent may dictate rule so that it is not new for Teague purposes), cert, denied, 523 U. S. 1088 (1998), or when lower courts have themselves adopted new rules and then determined that the Teague retroactivity factors apply, see Smith v. Groose, 205 F. 3d 1045, 1054 (CA8) (holding that Circuit rule that prosecution’s use of contradictory theories violates due process would fall within Teague’s “watershed” exception), cert, denied sub nom. Gammon v. Smith, 531 U. S. 985 (2000); Sanders v. Sullivan, 900 F. 2d 601, 606-607 (CA2 1990) (same, with respect to Circuit rule that prosecution’s unknowing use of material, perjured testimony violates Constitution). Here, consistent with such a purpose, the Supreme Court has previously spoken.

*677Second, the most likely consequence of the majority’s holding is further procedural complexity. After today’s opinion, the only way in which this Court can make a rule such as Cage’s retroactive is to repeat its Sullivan reasoning in a case triggered by a prisoner’s filing a first habeas petition (a “second or successive” petition itself being barred by the provision here at issue) or in some other case that presents the issue in a posture that allows such language to have the status of a “holding.” Then, after the Court takes the case and says that it meant what it previously said, prisoners could file “second or successive” petitions to take advantage of the now-clearly-made-applicable new rule. We will be required to restate the obvious, case by case, even when we have explicitly said, but not “held,” that a new rule is retroactive. See, e. g., Penry v. Lynaugh, 492 U. S. 302, 330 (1989) (stating that, if Court were to hold that Eighth Amendment prohibits execution of persons with mental retardation, this rule would be retroactively applicable on collateral review).

Even this complex route will remain open only if the relevant statute of limitations is interpreted to permit its 1-year filing period to run from the time that this Court has “made” a new rule retroactive, not from the time it initially recognized that new right. See 28 U. S. C. § 2244(d)(1)(C) (1994 ed., Supp. V) (limitations period runs from “the date on which the constitutional right asserted was initially recognized by the Supreme Court, if the right has been newly recognized by the Supreme Court and made retroactively applicable to cases on collateral review”). Otherwise, the Court’s approach will generate not only complexity, along with its attendant risk of confusion, but also serious additional unfairness.

I do not understand the basis for the Court’s approach. I fear its consequences. For these reasons, with respect, I dissent.