OPINION

delivered the opinion of the Court

Appellant Hossein Bagheri was convicted by a jury of driving while intoxicated. Tex. Penal Code § 49.04. The court sentenced appellant to four months in jail, probated for 18 months, and a fine of $1500 plus court costs. On appeal, appellant argued that the trial court had erred by admitting “retrograde extrapolation” testimony to show that the results of an Intoxilyzer test administered more than an hour after his arrest proved that he was intoxicated' while he was driving. The Court of Appeals reversed, holding that the admission of the testimony was error. Bagheri v. State, 87 S.W.3d 657 (Tex.App.San Antonio 2002). Although the jury was instructed that intoxication could be proven under two theories,1 and the evidence at trial may have been sufficient to support one of those theories without the extrapolation testimony, the Court of Appeals held *757that the erroneous admission of the testimony was not harmless. The Court of Appeals reasoned that, because the court’s charge submitted both theories of intoxication to the jury together, allowing the jury to return a general verdict of guilty, it was impossible to determine which theory the jury relied on to convict appellant. Id. at 660.

We granted review to address the State’s contentions on the following grounds:

(1) The Court of Appeals erred by presuming harm because the court’s charge submitted both DWI theories in a single question instead of conducting a proper harm analysis.

(2) The Court of Appeals failed to conduct a proper harm analysis by ignoring the overwhelming evidence of appellant’s guilt, by failing to detail the reasons why the court found that harm did occur, and by relying on a presumption of harm in charging the jury on both DWI theories in a general question instead of special issues.

(3) The Court of Appeals opinion relies upon a precedent taken wholly from the civil law framework for submitting “special issues” on liability to the jury which is in direct contravention of Texas Code of Criminal Procedure article 37.07.

Because we believe that the Court of Appeals reached the correct conclusion in finding harmful error, and did not improperly presume harm, we will affirm.

Officer Randall McCumbers stopped appellant at 2:30 a.m. after observing him speeding and driving erratically. At trial McCumbers testified that he first noticed appellant’s Mercedes convertible traveling at approximately 70 m.p.h. on the highway. As he watched, the vehicle veered off the road and onto the shoulder twice, and then sped up and cut across three lanes of traffic to exit without signaling. Officer McCumbers pulled over the vehicle and approached appellant to ask for his driver’s license and insurance. McCumbers testified that appellant was “very cordial,” and told him that he did not have his license but that he knew his number, which he recited from memory. Nevertheless, Officer McCumbers noted that appellant had trouble finding his insurance, seemed somewhat confused, and had slurred speech and red, glassy eyes. He asked appellant to step out of the car, at which time he noticed the smell of alcoholic beverages. Furthermore, he testified that appellant stumbled getting out of the car and was somewhat unsteady on his feet.

Officer McCumbers, who had been trained in field sobriety testing, administered various tests to appellant, including the Rhomberg balance test, the walk-and-turn test, the horizontal-gaze-nystagmus test, and the one-leg stand. Based on appellant’s performance on these tests, Officer McCumbers formed the opinion that appellant was intoxicated, and placed him under arrest for driving while intoxicated. After handcuffing him and placing him in the back of a police car, McCumbers and another officer arranged for appellant’s car to be impounded. Because McCumbers did not know how to close the Mercedes’ convertible top, he asked appellant to show him how to do it. Appellant got out of the police car, showed the officers how to close the top, and then returned to the police car.

After securing appellant’s car, Officer McCumbers transported him to the city jail Intoxilyzer room, where appellant agreed to give a voluntary breath sample. McCumbers, a certified Intoxilyzer operator, administered the Intoxilyzer test himself, at approximately 3:45 a.m. According to his testimony at trial, by the conclusion of this process Officer McCumbers had *758formed the opinion that appellant had lost the normal use of his mental and physical faculties.

The State called the breath test technical supervisor for Bexar County, A1 McDougall, to testify as an expert about the Intoxilyzer test. McDougall explained the scientific principles underlying the test and testified at length about the measures taken to ensure accurate test results. Over defense objection, the court ruled that the results of appellant’s Intoxilyzer test were admissible. Those results, showing that appellant’s alcohol concentration was greater than 0.10 slightly more than an hour after he was stopped, were brought before the jury.

McDougall then testified, again over defense objection, that studies involving alcohol consumption and driving indicated that almost everyone’s driving abilities are impaired when their alcohol concentration reaches 0.08. He also testified that in his own opinion, based on his personal observation of intoxicated persons, every person is impaired by the time they reach 0.08 alcohol concentration.

The State then began to question McDougall about the possibility, given a known breath alcohol concentration, of extrapolating backwards to determine what the breath alcohol concentration would have been at an earlier point in time (“retrograde extrapolation”). McDougall testified that such extrapolation is possible, if certain facts about the individual in question and about that individual’s drinking are known to the person performing the extrapolation. When the State questioned McDougall about average absorption and elimination rates for alcohol, the defense objected and asked for a hearing outside the presence of the jury to determine whether such testimony was based on scientifically rehable studies. See Kelly v. State, 824 S.W.2d 568 (Tex.Crim.App. 1992).2 During this hearing McDougall testified that, while elimination rates tend to be constant, and therefore predictable, absorption rates vary in- an individual based on a number of factors, rendering any attempt to define an “average” absorption rate meaningless. Furthermore, when a single breath test is used to measure alcohol concentration it is not evident whether the individual is in the absorption phase (in which case their alcohol concentration will be increasing), or in the elimination phase (in which case their alcohol concentration will be decreasing) at the time the test is administered. Therefore, without knowing a great deal of information about the individual and the circumstances of his or her drinking, it would not be possible to accurately extrapolate alcohol concentration based on a single breath test. However, McDougall testified that a study of drivers who were actually stopped on suspicion of driving while intoxicated (the “Goldberg” study), indicated that 90 to 95 percent of those drivers were in the elimination phase, rather than in the absorption phase. On cross-examination the defense tried to show that the Goldberg study was not rehable. Therefore, the defense argued, no assumption could be made about which phase, absorption or elimination, appehant was in when the In-toxilyzer test was administered, and no accurate range of alcohol concentration could be extrapolated from the results of that single test. The trial court disagreed and ruled the extrapolation testimony admissible.

McDougall testified based on “hypothetical” Intoxilyzer readings of 0.107 and *7590.113,3 that a subject’s alcohol concentration one hour earlier could have been between 0.107 and 0.143. On cross-examination he conceded that in order to come up with those numbers he had made certain assumptions about the hypothetical subject, including how many drinks he or she had consumed before stopping drinking, and how rapidly he or she had been drinking. In response to further defense questioning McDougall also admitted that the extrapolated range of alcohol concentrations could vary dramatically depending on several factors, including whether the subject was in the absorption phase or the elimination phase of alcohol metabolism. However, he adhered to his estimate of an alcohol concentration somewhere between 0.107 and 0.143 at the time of driving.

The State rested after McDougall’s testimony, and appellant took the stand in his own defense. He testified that he was born in Iran, and had moved to the United States at the age of eighteen.4 He described how he had worked busing and waiting tables while attending college in the United States, how he married and eventually became an American citizen, and how he and his brother currently owned several restaurants in the San Antonio area, as well three tanning salons and a day spa. On the day of his arrest appellant had been working at one of the restaurants from 7:30 the previous morning until 1:30 that morning. He had conducted an inventory after the restaurant closed, helped by a friend who worked at one of the other restaurants. Appellant testified that after completing the inventory, he gave his friend a ride to North Star Mall before heading home on highway 281. He indicated that he was very tired after working for so long in the restaurant, and that he habitually drove fast. He did not remember crossing over the line onto the shoulder, but testified that it probably happened when he looked down to change CDs while he was driving. Furthermore, he testified that he had decided at the last minute to take a different route home, and had cut across the lanes to make the exit just before Officer McCumbers pulled him over.

On cross-examination appellant testified that he had three to four drinks total on the evening of the arrest, and that he had the last one just before leaving the restaurant at approximately 1:30 a.m. He had eaten a full meal at approximately 8:00 p.m., and had been “munching on things” throughout the evening. He explained his poor performance on the field sobriety tests partly as fatigue, and partly as the result of back and neck injuries from previous car accidents.

The defense called Babak Mir, the friend who had been with appellant on the evening of his arrest. He confirmed that the two had performed an inventory at the restaurant, and that appellant had given him a ride to North Star Mall afterwards. He also testified that appellant did not appear intoxicated, and had not, in his opinion, lost the normal use of his mental and physical faculties. After Mir’s testimony, the defense rested.

The court then instructed the jury that a person is deemed to be intoxicated when he does not have the normal use of his mental or physical faculties, or when he *760has an alcohol concentration of 0.10 or more. During closing arguments, the State reiterated Officer McCumbers’ testimony that appellant appeared to be intoxicated based on the field sobriety tests. The State also emphasized McDougaU’s testimony regarding the retrograde extrapolation, and argued that this proved conclusively that appellant was “per se” intoxicated at the time he was stopped. The defense, in closing, argued that the results of the field sobriety tests were subjective, and that appellant actually performed fairly well on them. Furthermore, the defense argued that the results of the Intoxilyzer test were inaccurate, and that the extrapolation testimony was unreliable because the State’s expert had not considered variables which could have affected the results. On rebuttal, the State stressed the fact that Officer McCumbers’ personal observations supported the results of the scientific test. The jury returned a verdict of guilty.

One of the defense’s arguments on appeal was that the retrograde estimate of alcohol concentration should not have been admitted because the State did not provide a sufficient showing of scientific reliability under Kelly and under Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1998). Based on a case that this Court handed down while the appeal was being briefed, the State conceded error, but argued that it was harmless. Mata v. State, 46 S.W.3d 902 (Tex.Crim.App. 2001). Because a conviction could be sustained on either of the two alternate theories of intoxication, and evidence pertaining to each theory would support the other, the State argued that the jury’s general verdict should be upheld. The Court of Appeals held that the error was not harmless, and that the sufficiency of the evidence to convict under the alternate (impairment) theory was irrelevant, because the erroneously admitted evidence may have influenced the jury in their decision. Bagheri, 87 S.W.3d at 659-661. The Court of Appeals reversed the trial court judgment and remanded for a new trial.

On petition to this Court, the State argues that the Court of Appeals faded to perform a proper harm analysis, and instead “presumed harm” because the court’s charge submitted both theories of intoxication in a single question. The State contends that the Court of Appeals ignored “overwhelming evidence of guilt.” Moreover, according to the State, the Court of Appeals improperly borrowed a presumption of harm from a civil case, Crown Life Insurance Company v. Casteel, 22 S.W.3d 378 (Tex. 2000). Finally, the State notes that the Texas Legislature has mandated the admissibility of breath tests in DWI prosecutions. Tex. TRAnsp. Code Ann. § 724.064. Because the offense of DWI is defined as being intoxicated “while operating a motor vehicle” (Tex. Penal Code § 49.04), and because breath tests are always taken some time after arrest, the State argues that the Texas Legislature has effectively mandated that the jury engage in retrograde extrapolation.

Addressing the last of these arguments first, we do not interpret § 724.064 of the Transportation Code as a legislative mandate requiring juries to engage in retrograde extrapolation. The determination that breath test results should be admissible in DWI prosecutions without the necessity of establishing the scientific basis for such tests under Daubert and Kelly does not, in our opinion, relieve the State of the burden of showing that those test results in any given case are relevant, in the sense that they accurately reflect the subject’s alcohol concentration at the time of the offense. As this Court explained *761when the “per se” definition of intoxication was first added to the DWI statute:

To prove the element of intoxication in a prosecution for the offense of driving while intoxicated, the State must offer proof beyond a reasonable doubt as to that element. To be sure, if the State relies upon the 0.10 % definition of intoxication, then such proof will normally appear in the form of a chemical test showing the alcohol concentration in a defendant’s body near the time of the offense. However, a conviction will not necessarily follow from the offer of such a test. First, the trier of fact must still be convinced beyond a reasonable doubt that the chemical test provides trustworthy evidence of alcohol concentration in a defendant’s breath, blood or urine. Second, the jury must still be convinced beyond a reasonable doubt that an inference can be made from the results of the chemical test that the defendant had a 0.10 % alcohol concentration in his body at the time of the offense.

(Emphasis in original) Forte v. State, 707 S.W.2d 89, 94-95 (Tex.Crim.App. 1986).

At trial, on appeal, and on petition for discretionary review to this Court, there has been much discussion about whether the erroneous submission of the “per se theory” of intoxication necessarily requires reversal if there was sufficient evidence to convict under the “impairment theory.” In the past, the San Antonio Court of Appeals had held that the submission of an “invalid theory” along with a “valid theory” in a jury charge is harmless if there is sufficient evidence to convict on the valid theory. See, e.g., Mata v. State, 75 S.W.3d 499, 501-502 (Tex.App.-San Antonio 2002); Hartman v. State, 2 S.W.3d 490, 494 (Tex. App.-San Antonio 1999, pet. ref'd). On the other hand, the Court of Appeals in this case cited to Casteel, a civil case in which the Texas Supreme Court held:

[W]hen a jury bases a finding of liability on a single broad-form question that commingles invalid theories of liability with valid theories, the appellate court is often unable to determine the effect of this error. The best the court can do is determine that some evidence could have supported the jury’s conclusion on a legally valid theory.... Accordingly, we hold that when a trial court submits a single broad-form liability question incorporating multiple theories of liability, the error is harmful and a new trial is required when the appellate court cannot determine whether the jury based its verdict on an improperly submitted invalid theory.

Casteel, 22 S.W.3d at 388.

The State is correct in its contention that Casteel is not applicable here.5 In *762that case, four out of five of the grounds for liability submitted to the jury were invalid because the litigant did not have standing to bring them. Id. Each of the grounds alleged independent theories of liability, involving various acts of misrepresentation and failure to disclose information. Id. at 386-388. In this case, by contrast, the two “theories” of intoxication do not involve separate violations of the law. Several Courts of Appeals have concluded that “[t]he definition of ‘intoxicated’ in the DWI statute sets forth alternate means of committing one offense. It does not set forth separate and distinct offenses.” Ex Parte Crenshaw, 25 S.W.3d 761 (Tex.App. Houston 1st Dist.2000, pet. refd); see also, Kilgo v. State, 880 S.W.2d 828, 829 (Tex.App.-Dallas 1994, pet. ref'd); Harris v. State, 866 S.W.2d 316, 324 (Tex.App.-San Antonio 1993, pet. ref'd).

We will go further, and say that the definitions contained in § 49.01 set forth alternate means by which the State may prove intoxication, rather than alternate means of committing the offense. The conduct proscribed by the Penal Code is the act of driving while in a state of intoxication. That does not change whether the State uses the per se definition or the impairment definition to prove the offense. Therefore, the harm analysis in this case should be performed as it would be in any case where expert testimony was erroneously admitted, without regard to the “validity” of either theory of intoxication.

However, we do not agree that the Court of Appeals relied on Casteel to create an improper presumption of harm. Rather, we interpret the court’s reference to that case as a response to one of the State’s arguments on appeal. As the court explained:

Without knowing that the jury found Bagheri guilty under the impairment theory, the State’s assertion that the evidence is sufficient to support conviction under that theory is simply irrelevant; it in no way leads us to a “fair assurance that the error [in admitting the retrograde extrapolation evidence] did not influence the jury, or had but a slight effect.” [citation omitted]. We therefore hold the sufficiency of the evidence to support conviction under an impairment theory does not alone establish that the error in admitting a flawed retrograde extrapolation analysis was harmless... .We will instead adhere to the harm test enunciated by the Texas Court of Criminal Appeals [in Solomon ].

Bagheri, 87 S.W.3d at 660-661, quoting Solomon v. State, 49 S.W.3d 356, 365 (Tex.Crim.App. 2001).

The erroneous admission of the extrapolation testimony in this case did *763not rise to the level of constitutional error. See Potier v. State, 68 S.W.3d 657, 662-663 (Tex.Crim.App. 2002)(“Erroneous evidentiary rulings rarely rise to the level of denying the fundamental constitutional rights ... ”). In considering non-constitutional error, an appellate court must disregard the error if the court, “after examining the record as a whole, has fair assurance that the error did not influence the jury, or had but a slight effect.” Solomon 49 S.W.3d at 365; Tex.R.App. P. 44.2(b). The question is not whether there was sufficient evidence to support the verdict. Instead, the reviewing court should consider the entire record when making this determination, including testimony, physical evidence, jury instructions, the State’s theories and any defensive theories, closing arguments, and voir dire if applicable. Motilla v. State, 78 S.W.3d 352, 355 (Tex.Crim.App. 2002). Important factors are “the nature of the evidence supporting the verdict, the character of the alleged error and how it might be considered in connection with other evidence in the case.” Id. More specifically, the reviewing court should consider whether the State emphasized the error, whether the erroneously admitted evidence was cumulative, and whether it was elicited from an expert. Id. at 356; Solomon, 49 S.W.3d at 365.

Given our view that the two “theories” of intoxication do not define separate offenses, we believe that the Court of Appeals’ reasoning, although flawed, nevertheless led to the correct result. The court noted that the jury’s general verdict in this case made it impossible to determine which “theory” the jury relied upon. Bagheri, 87 S.W.3d at 660. We believe that this fact alone is not determinative, because evidence to prove intoxication under either definition is relevant to the single question of whether appellant was, in fact intoxicated. The issue was whether the erroneously admitted testimony might have prejudiced the jury’s consideration of the other evidence or substantially affected their deliberations.

An examination of the Motilla and Solomon factors leads us to the conclusion that the Court of Appeals was correct in holding that the erroneous admission of the retrograde extrapolation testimony was not harmless. The testimony in question was “elicited from an expert”. Solomon, 49 S.W.3d at 365. Throughout the trial, from voir dire to closing arguments, the State emphasized the fact that there would be scientific evidence from an expert showing that appellant’s alcohol concentration was greater than 0.10 at the time of operating the vehicle. During direct examination, the State inquired in great detail about McDougall’s qualifications, and McDougall offered his “expert opinion” on a number of subjects. He testified (over defense objection), that everyone is impaired at 0.08 alcohol concentration. He offered his opinion, as an expert, that the scientific theory behind retrograde extrapolation was reliable. Although he acknowledged that an exact concentration could not be calculated without knowing more about the subject, he testified that he could still extrapolate and determine a range of alcohol concentration “within a reliable range.” In closing argument, the State emphasized that McDougall had studied with “the leading experts in the field in the United States,” and that he had calculated the extrapolated range based on “scientifically rehable, valid evidence.” Clearly, under Motilla and Solomon, the erroneously admitted evidence was both elicited from an expert and emphasized by the State.

Those cases also call for an examination of whether the evidence in question was “cumulative.” Motilla, 78 S.W.3d at 356; *764Solomon, 49 S.W.3d at 365. We do not believe that the extrapolation testimony in this ease was cumulative. The State’s claim that there was “overwhelming evidence of guilt” is something of an exaggeration. Although there was testimony from Officer McCumbers about appellant’s apparent intoxication and his poor performance on the field sobriety tests, this testimony was, as the defense pointed out, somewhat subjective. The appellant admitted to drinking alcoholic beverages earlier in the evening, but presented testimony supporting his assertion that he was not intoxicated at the time he was stopped. There was some testimony indicating that his erratic driving and poor performance on the sobriety tests were caused by fatigue. While the jury was free to disbelieve this testimony, the effect of the “scientifically reliable” extrapolation evidence was almost certainly to tip the balance in favor of the State. During voir dire, numerous members of the jury pool expressed a belief that a person who fails a breath test is “flat-out” guilty of DWI. Several venire members also believed that alcohol concentration would necessarily decline over time, and that a subject’s alcohol concentration would therefore always be higher at the time of driving than it would be on a later Intoxilyzer test. While this information is not dispositive, it is an indication of the powerful persuasive effect that “scientific” evidence has on the average juror.

After a thorough examination of the entire record in this case, we agree with the Court of Appeals. We cannot say with fair assurance that the erroneous admission of McDougall’s retrograde extrapolation testimony did not influence the jury, or had but a slight effect. Therefore, the judgment of the Court of Appeals is affirmed.

dissenting in which KEASLER, J., joined.

I respectfully dissent. In cases like this, I agree with the Court’s decision that sufficiency of the evidence to support a conviction under an “impairment” theory of intoxication is relevant to, but not necessarily dispositive of, the Rule 44.2(b) harm analysis for the erroneous admission of retrograde analysis evidence. See Bagheri v. State, 119 S.W.3d at 763 (Tex.Cr.App., No. 1251-02, delivered this date) (“question is not whether there was sufficient evidence to support the verdict”) and 119 S.W.3d at 763 (setting out various factors, including the entire record and the testimony, for appellate courts to consider in determining whether erroneous admission of evidence “influence[d] the jury, or had but a slight effect”); compare Griffin v. United States, 502 U.S. 46, 112 S.Ct. 466, 469-74, 116 L.Ed.2d 371 (1991); Kitchens v. State, 823 S.W.2d 256, 259 (Tex.Cr.App. 1991), cert. denied, 504 U.S. 958, 112 S.Ct. 2309, 119 L.Ed.2d 230 (1992). The Court of Appeals, therefore, erred to decide that sufficiency of the evidence to support a conviction under an “impairment” theory is “simply irrelevant” to this harm analysis. See Bagheri v. State, 87 S.W.3d 657, 660 (Tex.App.-San Antonio 2002).1

Having decided that the Court of Appeals applied an improper Rule 44.2(b) *765harm analysis, the Court’s opinion, rather than remanding the ease to the Court of Appeals, conducts its own Rule 44.2(b) harm analysis and concludes that the error was not harmless. See Bagheri, 119 S.W.3d at 763-64. I disagree.

The error in this case occurred when McDougall was permitted to testify that, based on “hypothetical” intoxilyzer readings of 0.107 and 0.113 (which happened to be appellant’s intoxilyzer readings about an hour after the police stopped him), a subject’s alcohol concentration an hour earlier could have been between 0.107 and 0.143.2 See Bagheri, 119 S.W.3d at 758-59; Bagheri, 87 S.W.3d at 659. On cross-examination, however, McDougall qualified this with testimony that the “extrapolated range of alcohol concentrations could vary dramatically depending on several factors” that he apparently had not considered See Bagheñ, 119 S.W.3d at 758-59, 760. The defense reminded the jury of this during closing arguments when the defense stated “that the extrapolation testimony was unreliable because the State’s expert had not considered variables which could have affected the results.” See Bagheri, 119 S.W.3d at 760. During its closing arguments, the State equally emphasized evidence supporting both theories of intoxication. See Bagheri, 119 S.W.3d at 759-60.

Other properly admitted evidence proved that appellant’s intoxilyzer readings about an hour after the police stopped him were 0.107 and 0.113. See Bagheri, 87 S.W.3d at 659. There was other properly admitted evidence that “almost everyone’s driving abilities are impaired when their alcohol concentration reaches 0.08.” See Bagheri, 119 S.W.3d at 758. There was a wealth of evidence that appellant was intoxicated and impaired when he was stopped.3 This also constitutes at least some evidence that even supports the “per se” theory of intoxication. I would hold that, when a record such as the one here contains a wealth of evidence to support the “impairment” theory and some evidence to support the “per se” theory, any error in admitting somewhat qualified retrograde analysis evidence like that here “did not influence the jury, or had but a slight effect.”

Even though cases such as Griffin and Kitchens are not necessarily dispositive of the harm analysis, they nevertheless do have something relevant to say about it:

It is one thing to negate a verdict that, while supported by evidence, may have been based on an erroneous view of the law; it is another to do so merely on the chance-remote, it seems to us-that the jury convicted on a ground that was not supported by adequate evidence when there existed alternative grounds for which the evidence was sufficient. (Citation and internal quotes omitted).

Griffin, 112 S.Ct. at 474.

I respectfully dissent.