A conviction based on a general verdict is subject to challenge if the jury was instructed on alternative theories of guilt and may have relied on an invalid one. See Stromberg v. California, 283 U. S. 359 (1931); Yates v. United States, 354 U. S. 298 (1957). In this case the Court of Appeals for the Ninth Circuit held that such an error is “structural error,” requiring that the conviction be set aside on collateral review without regard to whether the flaw in the instructions prejudiced the defendant. The parties now agree that the Court of Appeals was wrong to categorize this type of error as “structural.” They further agree that a reviewing court finding such error should ask whether the flaw in the instructions “had substantial and injurious effect or influence in determining the jury’s verdict.” Brecht v. Abrahamson, 507 U. S. 619, 623 (1993) (internal quotation marks omitted). We agree as well and so hold.
*59Respondent Michael Pulido was convicted by a California jury of felony murder. On direct appeal, Pulido sought to vacate his conviction on the ground that the jury instructions were erroneous: They permitted the jury to find him guilty of felony murder if he formed the intent to aid and abet the underlying felony before the murder, but they also permitted the jury to find him guilty if he formed that intent only after the murder. The California Supreme Court agreed with Pulido that the latter theory was invalid under California law, but upheld the conviction on the ground that Pulido was not prejudiced by the error. People v. Pulido, 15 Cal. 4th 713, 727, 936 P. 2d 1235, 1243-1244 (1997). Pulido sought federal habeas relief, which the District Court granted after concluding that instructing the jury on the invalid theory had a “‘substantial and injurious effect or influence in determining the jury’s verdict.’” Pulido v. Lamarque, No. C 99-4933 CW (PR) (ND Cal., Mar. 24, 2005), App. to Pet. for Cert. 65a-66a (quoting Brecht, supra, at 637).
The State appealed, and the Court of Appeals affirmed. Pulido v. Chrones, 487 F. 3d 669 (2007) (per curiam). On appeal, Pulido argued the District Court’s Brecht analysis was correct, but in the alternative sought to avoid the harmless-error inquiry altogether. In support of that alternative argument, he maintained that when a jury returns a general verdict after being instructed on both a valid and an invalid theory, the conviction must be automatically set aside, without asking whether the invalid instruction was harmless. The Court of Appeals recognized that the Brecht “substantial and injurious effect” standard governs harmless-error analysis on federal habeas, 487 F. 3d, at 673, n. 3 (internal quotation marks omitted), but agreed with Pulido that instructing a jury on multiple theories of guilt, one of which is legally improper, was “structural” error exempting the instructions as a whole from harmless-error review, id., at 675-676. Such error instead required setting aside the conviction on habeas unless the reviewing court *60could determine with “ ‘absolute certainty’ ” that the defendant was convicted under a proper theory. Id., at 676 (quoting Lara v. Ryan, 455 F. 3d 1080, 1086 (CA9 2006)). Because the instructions “le[ft] open the possibility” that the jury convicted Pulido on the impermissible ground, the court concluded that the verdict must be reversed. 487 F. 3d, at 676. We granted certiorari. Chrones v. Pulido, 552 U. S. 1230 (2008).
The Ninth Circuit precedent on which the Court of Appeals relied, see Lara v. Ryan, supra, based its structural-error analysis upon a line of our cases beginning with Stromberg. Stromberg addressed the validity of a general verdict that rested on an instruction that the petitioner could be found guilty for displaying a red flag as “‘a sign, symbol, or emblem of opposition to organized government, or [a]s an invitation or stimulus to anarchistic action, or [a]s [a]n aid to propaganda that is of a seditious character.’ ” 283 U. S., at 363. After holding that the first clause of the instruction proscribed constitutionally protected conduct, we concluded that the petitioner’s conviction must be reversed because “it [wa]s impossible to say under which clause of the [instruction] the conviction was obtained.” Id., at 368. In Yates v. United States, supra, we extended this reasoning to a conviction resting on multiple theories of guilt when one of those theories is not unconstitutional, but is otherwise legally flawed.
Both Stromberg and Yates were decided before we concluded in Chapman v. California, 386 U. S. 18 (1967), that constitutional errors can be harmless. Accordingly, neither Stromberg nor Yates had reason to address whether the instructional errors they identified could be reviewed for harmlessness, or instead required automatic reversal. In a series of post -Chapman cases, however, we concluded that various forms of instructional error are not structural but instead trial errors subject to harmless-error review. See, e. g., Neder v. United States, 527 U. S. 1 (1999) (omission of *61an element of an offense);. California v. Roy, 519 U. S. 2 (1996) (per curiam) (erroneous aider and abettor instruction); Pope v. Illinois, 481 U. S. 497 (1987) (misstatement of an element of an offense); Rose v. Clark, 478 U. S. 570 (1986) (erroneous burden shifting as to an element of an offense).
Although these cases did not arise in the context of a jury instructed on multiple theories of guilt, one of which is improper, nothing in them suggests that a different harmless-error analysis should govern in that particular context. To the contrary, we emphasized in Rose that “while there are some errors to which [harmless-error analysis] does not apply, they are the exception and not the rule.” Id., at 578. And Neder makes clear that harmless-error analysis applies to instructional errors so long as the error at issue does not categorically “ ‘vitiat[e] all the jury’s findings.’” 527 U. S., at 11 (quoting Sullivan v. Louisiana, 508 U. S. 275, 281 (1993) (erroneous reasonable-doubt instructions constitute structural error)). Aminstructional error arising in the context of multiple theories of guilt no more vitiates all the jury’s findings than does omission or misstatement of an element of the offense when only one theory is submitted.
In fact, drawing a distinction between alternative-theory error and the instructional errors in Neder, Roy, Pope, and Rose would be “patently illogical,” given that such a distinction “ ‘reduces to the strange claim that, because the jury ... received both a “good” charge and a “bad” charge on the issue, the error was somehow more pernicious than . . . where the only charge on the critical issue was a mistaken one.’ ” 487 F. 3d, at 677-678 (O’Scannlain, J., concurring specially) (quoting Quigley v. Vose, 834 F. 2d 14, 16 (CA1 1987) (per curiam)); see also Becht v. United States, 403 F. 3d 541, 548 (CA8 2005) (same), cert. denied, 546 U. S. 1177 (2006).
Pulido now agrees with the State that the Court of Appeals erred by treating the instructional error in this case as structural, and that the required prejudice analysis should *62be governed by Brecht’s “substantial and injurious effect” standard. See Brief for Respondent 17 (“[T]he Ninth Circuit was mistaken in its ‘structural defect’ nomenclature”); Tr. of Oral Arg. 27 (“We acknowledge that this is a trial error and is subject to the Brecht prejudice standard”). So do we.
Pulido nonetheless maintains we should affirm because the Court of Appeals effectively engaged in the Brecht analysis, despite its clear description of the error as “structural.” But despite full briefing on the applicability of Brecht, the Court of Appeals mentioned Brecht only briefly in a footnote, see 487 F. 3d, at 673, n. 3, and then went on to agree with Pulido’s alternative assertion that “the instructional error was structural and therefore not subject to harmless error review,” id., at 675-676. The court also stated that the conviction had to be overturned unless the court was “‘absolutely certain’ ” that the jury relied on a valid ground. Id., at 676. Such a determination would appear to be a finding that no violation had occurred at all, rather than that any error was harmless. In any event, an “absolute certainty” standard is plainly inconsistent with Brecht. Accordingly, we express no view on whether Pulido is entitled to habeas relief, but rather remand to the Court of Appeals for application of Brecht in the first instance.*
The judgment is vacated, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.
The dissent maintains the Court of Appeals “undertook a searching review of the parties’ evidence and the jury instructions to determine the error’s effect on the jury.” Post, at 68 (opinion of Stevens, J.). But the Court of Appeals reached its conclusion based solely on the existence of a typographical error in the special circumstances instructions, without addressing any of the State’s arguments that the typographical error was harmless in light of the record as a whole. There was no need for that court to address those arguments, of course, because of its mistaken conclusion that the instructional error was structural. Under such circumstances, remand is the appropriate course.