with whom Justice Souter and Justice Ginsburg join, dissenting.
The Court of Appeals misused the term “structural error” in its opinion affirming the District Court’s order granting Pulido’s application for a writ of habeas corpus. But the court’s misnomer was inconsequential because its decision rested on substantially the same analysis as the District Court’s, which correctly applied the standards set forth in Kotteakos v. United States, 328 U. S. 750 (1946), Brecht v. Abrahamson, 507 U. S. 619 (1993), and O’Neal v. McAninch, 513 U. S. 432 (1995). The Court of Appeals’ decision therefore did not warrant this Court’s review and does not now merit a remand to require that court to repeat its analysis. In my opinion, the interest in expediting the conclusion of this protracted litigation outweighs the interest in correcting a misnomer.
Respondent Michael Pulido was charged with felony murder for robbing a gas station and killing the attendant. At trial, the State argued that Pulido acted alone. Pulido maintained that his uncle was the principal actor and that he had no knowledge of his uncle’s plan when the two arrived at the gas station. While he was waiting in the car, Pulido claimed, he heard a shot and ran into the store. At that point, his uncle insisted that Pulido help him pry open the stolen cash register and dispose of it, and Pulido reluctantly complied. The jury convicted Pulido of felony murder, but it was unable to reach a verdict on the charges that Pulido personally used a firearm and intentionally inflicted great bodily harm.
As a matter of California law, felony-murder liability extends to all persons jointly engaged in the commission of a felony at the time of a killing when one of the joint actors kills in furtherance of the common design. People v. Pulido, 15 Cal. 4th 713, 716, 936 P. 2d 1235, 1236 (1997). But a person is not guilty of felony murder if he is only a “late-joining” aider and abettor — i. e., if he did not himself commit the mur*64der and his participation in the underlying felony did not begin until after the victim was killed. Ibid. In this case, the trial court’s instructions erroneously permitted the jury to find Pulido guilty on such a theory, as they did not require the jury to find either that Pulido committed the murder or that he aided and abetted the underlying robbery before the murder was committed. Because the instructions allowed the jury to convict Pulido of felony murder for conduct that does not amount to that offense, their inclusion was constitutional error.
On direct appeal, the California Supreme Court agreed with Pulido that the late-joiner theory was an invalid theory of felony-murder liability. Ibid. It nevertheless held that any error in the trial court’s instructions was harmless. According to the court, the jury found that Pulido had been engaged in the robbery at the time of the killing because the robbery-murder special-circumstance instruction stated that “the murder was committed while the defendant was engaged” in the “commission of or attempted commission of [a] robbery.” Id., at 727, 936 P. 2d, at 1243 (citing Cal. Penal Code Ann. § 190.2(a)(17) (internal quotation marks omitted)). Based on that portion of the instruction, the court concluded that the special-circumstance verdict “demonstrates the jury did not accept the theory [Pulido] joined the robbery only after [the victim] was killed,” and it therefore held that Pulido was not prejudiced by the error. 15 Cal. 4th, at 727, 936 P. 2d, at 1244.
In reaching that conclusion, however, the California Supreme Court failed to take into account the entire special-circumstance instruction. A typographical error in that instruction in fact permitted the jury to find the special circumstance of robbery-murder true if it found either that the murder was committed while the defendant was engaged in the commission of a robbery or that it “was committed in order to carry out or advance the commission of the crime of robbery or to facilitate the escape therefrom or to avoid *65detection” — a finding entirely consistent with the late-joiner theory. App. 14. Thus, as the State concedes, the erroneous instructions made it “reasonably likely” that the jury convicted Pulido on the impermissible theory. Brief for Petitioner 18.
After exhausting his state postconviction remedies,1 Pulido sought a writ of habeas corpus in Federal District Court. The District Court recognized the erroneous disjunctive in the special-circumstance instruction that the California Supreme Court had overlooked, and it held that the state court’s conclusion that Pulido was not prejudiced by erroneous instructions was “an objectively unreasonable application of clearly established federal law.” Pulido v. Lamarque, No. C 99-4933 CW (PR) (ND Cal., Mar. 24, 2005), App. to Pet. for Cert. 64a.
The District Court then considered the effect of that error on the jury. Correctly relying on Brecht, the District Court began its analysis by noting that a federal habeas petitioner “is not entitled to habeas relief unless the State court’s error resulted in actual prejudice, that is, the error had a ‘substantial and injurious effect or influence in determining the jury’s verdict.’” App. to Pet. for Cert. 65a (quoting Brecht, 507 U. S., at 637). If an error had a substantial influence, or if “the record is so evenly balanced that a conscientious judge is in ‘grave doubt’ ” as to whether it had such an effect, the conviction must be reversed. App. to Pet. for Cert. 65a (quoting O’Neal, 513 U. S., at 438).
To determine whether the error was harmless under this standard, the District Court scrutinized the record, including the arguments of both parties, the evidence supporting their respective theories of the case, the jury instructions, the jury’s questions to the trial court, and the various parts of the jury’s verdict. App. to Pet. for Cert. 65a-66a. *66Throughout this inquiry, the District Court properly avoided substituting its judgment for the jury’s. As we cautioned in Kotteakos, in undertaking harmless-error analysis “it is not the [reviewing] court’s function to determine guilt or innocence. Nor is it to speculate upon probable reconviction and decide according to how the speculation comes out.” 328 U. S., at 763 (citations omitted). Thus, “[t]he inquiry cannot be merely whether there was enough to support the result” in the absence of the error. Id., at 765. Rather, the proper question is “whether the error itself had substantial influence. If so, or if one is left in grave doubt, the conviction cannot stand.” Ibid.; accord, O’Neal, 513 U. S., at 437.
That was precisely the question addressed by the District Court when it sought to ascertain what the jury actually found. The court concluded that, while it is “possible” that the jury found that Pulido aided and abetted the robbery before the victim was killed, the court had “no way of determining whether this was the case.” App. to Pet. for Cert. 66a. Because that uncertainty left the court with “ ‘grave doubt as to the likely effect of [the] error on the jury’s verdict,’ ” it faithfully applied the standard mandated by Kotteakos and O’Neal and found that the error was not harmless. App. to Pet. for Cert. 66a (quoting O’Neal, 513 U. S., at 435).
On appeal, Pulido contended that the judgment of the District Court should be affirmed whether the instructional error was viewed as structural error or as trial error. Brief for Appellee and Cross-Appellant in No. 05-15916 etc. (CA9), pp. 53-64 (hereinafter Appellee’s Brief). He argued that the error was not harmless under Brecht and O’Neal because the substantial evidence that supported the invalid theory made it likely that the jury convicted him on that basis. Appellee’s Brief 55-64. In particular, Pulido noted that the “injurious effect” of this type of error “is greatest when the instruction compromises the defense by appearing to extend liability even to .the factual scenario suggested by the de*67fense evidence,” as was true in this case. Id., at 57 (internal quotation marks omitted; emphasis deleted). At oral argument, the parties’ contentions similarly focused on the Brecht standard and the result that harmless-error ánalysis required.
Less than two months after oral argument, and before the Court of Appeals issued its decision in this case, a different panel of the Ninth Circuit decided Lara v. Ryan, 455 F. 3d 1080 (2006). Lara was convicted of attempted murder by a jury that had been instructed that it could find him guilty under either an express malice theory or an implied malice theory, the second of which is legally invalid. Id., at 1082. The Ninth Circuit described the erroneous instruction as a “structural error,” but it held that such an error does not necessitate reversal when a reviewing court can “determine with absolute certainty” that the defendant was not convicted under the erroneous theory. Id., at 1086. Because the jury “made a specific finding that Lara attempted to murder willfully, deliberately, and with premeditation,” the court concluded that it necessarily relied on the valid instruction and that reversal was therefore not required. Id., at 1086-1087.
In those limited instances in which this Court has found an error “structural,” we have done so because the error defies analysis by harmless-error standards. See Arizona v. Fulminante, 499 U. S. 279, 309 (1991); see also United States v. Gonzalez-Lopez, 548 U. S. 140, 150 (2006) (quoting Sullivan v. Louisiana, 508 U. S. 275, 282 (1993); United States v. Cronic, 466 U. S. 648, 659, and n. 25 (1984)). Indeed, it is because the consequences “‘are necessarily unquantifiable and indeterminate’” that automatic reversal is required when such errors occur. Gonzalez-Lopez, 548 U. S., at 150 (quoting Sullivan, 508 U. S., at 282). That the court in Lara could be “ ‘ “absolutely certain” that the jury relied upon the legally correct theory,’” 455 F. 3d, at 1085, *68shows both that the error was susceptible of harmless-error analysis and that the court in fact found the error harmless despite repeatedly referring to it as “structural.”
Citing Lara, the Court of Appeals’ per curiam opinion labeled the erroneous instruction in this case a structural error.2 As in Lara, the court then undertook a searching review of the parties’ evidence and the jury instructions to determine the error’s effect on the jury. Noting, among other things, that “[t]he typographical error in the contemporaneity instruction relied upon by the California Supreme Court introduces doubt into any inference to be drawn from the jury’s finding as to the special circumstance,” the court concluded that “the jury instructions leave open the possibility that the jury convicted Pulido on a legally impermissible theory.” Pulido v. Chrones, 487 F. 3d 669, 676 (CA9 2007) (per curiam). That possibility of reliance on the erroneous instruction is the “substantial and injurious effect” to which Brecht refers. Thus, although the Court of Appeals called the error in this case by the wrong name, it performed substantially the same analysis and reached the same conclusion as the District Court did when it applied the standard prescribed by Brecht.3
Judge Thomas concurred separately both to defend the Lara decision and to demonstrate that harmless-error analysis also supports the panel’s result. 487 F. 3d, at 678-683. Unlike the District Court, Judge Thomas applied the *69harmless-error standard announced in Chapman v. California, 386 U. S. 18 (1967), instead of looking to Brecht. 487 F. 3d, at 678. But his analysis similarly establishes that at least some jurors very likely relied on the impermissible late-joiner theory. Id., at 679-683.
The record before us clearly supports that conclusion. Indeed, even petitioner admits that the ambiguity in the robbery and murder instructions and the trial court’s confusing answers to the jury’s questions “combined to make it reasonably likely that the jury applied the instructions in an unconstitutional way.” Brief for Petitioner 18. That reasonable likelihood is sufficient to support the conclusion that the error was not harmless under Brecht.
Because the District Court’s analysis was correct and the Court of Appeals’ result was substantially the same, I think this Court’s decision to remand for the purpose of obtaining a third analysis of the harmless-error issue is a misuse of scarce judicial resources. I would therefore affirm the judgment of the Court of Appeals.
The California Supreme Court summarily denied Pulido’s state petition for a writ of habeas corpus. See Pulido v. Chrones, 487 F. 3d 669, 673 (CA9 2007) (per curiam).
The Court of Appeals’ opinion asserts that Pulido argued that the error was structural under Lara v. Ryan, 455 F. 3d 1080 (CA9 2006). But due to the timing of the Lara decision, the parties did not raise arguments relying on that precedent until their postargument supplemental briefing. As discussed above, the parties’ arguments had initially focused on the proper application of Brecht v. Abrahamson, 507 U. S. 619 (1993).
The Court of Appeals in fact cited Brecht and recited the proper standard in a footnote before turning its attention to Lara: “If there is constitutional error, we consider whether the error was harmless; that is, whether the error had a ‘ “substantial and injurious effect or influence in determining the jury’s verdict.” ’ ” 487 F. 3d, at 673, n. 3.