A California court convicted the respondent Kenneth Roy of the robbery and first-degree murder of Archie Mannix. The State's theory, insofar as is relevant here, was that Roy, coming to the aid of a confederate who was trying to rob Mannix, helped the confederate kill Mannix. The trial judge gave the jury an instruction that permitted it to convict Roy of first-degree murder as long as it concluded that (among other things) Roy, "with knowledge of" the confederate's "unlawful purpose" (robbery), had helped the confederate, i. e., had "aid[ed]," "promote[d]," "encourage[d]," or "instigate[d]" by "act or advice the commission of" the confederate's crime. The California Supreme Court later held in People v. Beeman, 35 Cal. 3d 547, 561, 674 P. 2d 1318, 1326 (1984), that an identical instruction was erroneous because of what it did not say, namely that state law also required the jury to find that Roy had the "knowledge [and] intent or purpose of committing, encouraging, or facilitating" the confederate's crime. Id., at 561, 674 P. 2d, at 1326 (emphasis added). Despite this error, the California Court of Appeal affirmed Roy's conviction because it found the error "harmless beyond a reasonable doubt." See Chapman v. California, 386 U.S. 18, 24 (1967). The California Supreme Court denied post-conviction relief.
Subsequently Roy, pointing to the same instructional error, asked a Federal District Court to issue a writ of habeas corpus. The District Court denied the request because, in its view, the error was harmless. Indeed, the District Court wrote that no rational juror could have found that Roy knew the confederate's purpose and helped him but also found that Roy did not intend to help him. A divided Ninth Circuit panel affirmed. Roy v. Gomez, 55 F. 3d 1483 (1995).
The Ninth Circuit later heard the case en banc and reversed the district court. It held that the instructional error was not harmless. 81 F. 3d 863 (1996). In doing so, the majority applied a special "harmless error" standard, which it believed combined aspects of our decisions in Carella v. California, 491 U.S. 263 (1989), and O'Neal v. McAninch, 513 U. S. ___ (1995). The Ninth Circuit described the standard as follows:
We believe that the State, and the dissenting judges in the Ninth Circuit, are correct about the proper standard. The Ninth Circuit majority drew its special standard primarily from a concurring opinion in Carella, supra, a case that dealt with legal presumptions. The concurrence in that case set out the views of several Justices about the proper way to determine whether an error in respect to the use of a presumption was "harmless." Subsequent to Carella, however, this Court held that a federal court reviewing a state court determination in a habeas corpus proceeding, ordinarily should apply the "harmless error" standard that the Court had previously enunciated in Kotteakos v. United States, 328 U.S. 750 (1946), namely "whether the error `had substantial and injurious effect or influence in determining the jury's verdict.' " Brecht, supra, at 637 (citing Kotteakos v. United States, 328 U.S. 750, 776 (1946)). The Court recognized that the Kotteakos standard did not apply to " `structural defects in the constitution of the trial mechanism, which defy analysis by "harmless-error" standards,' " 507 U.S., at 629 , but held that the Kotteakos standard did apply to habeas review of what the Court called "trial errors," including errors in respect to which the Constitution requires state courts to apply a stricter, Chapman-type standard of "harmless error" when they review a conviction directly. 507 U.S., at 638 . In O'Neal, supra, this Court added that where a judge, in a habeas proceeding, applying this standard of harmless error, "is in grave doubt as to the harmlessness of an error," the habeas "petitioner must win." Id., at ___ (slip op., at 3).
The case before us is a case for application of the "harmless error" standard as enunciated in Brecht and O'Neal. This Court has written that "constitutional error" of the sort at issue in Carella is a "trial error," not a "structural error," and that it is subject to "harmless error" analysis. Arizona v. Fulminante, 499 U.S. 279, 306 -307 (1991). The state courts in this case applied harmless-error analysis of the strict variety, and they found the error "harmless beyond a reasonable doubt." Chapman, supra, at 24. The specific error at issue here-an error in the instruction that defined the crime-is, as the Ninth Circuit itself recognized, as easily characterized as a "misdescription of an element" of the crime, as it is characterized as an error of "omission." 81 F. 3d, at 867, n. 4. No one claims that the error at issue here is of the "structural" sort that " `def[ies] analysis by "harmless error" standards.' " Brecht, supra, at 629. The analysis advanced by the Ninth Circuit, while certainly consistent with the concurring opinion in Carella, does not, in our view, overcome the holding of Brecht, followed in O'Neal, that for reasons related to the special function of habeas courts, those courts must review such error (error that may require strict review of the Chapman-type on direct appeal) under the Kotteakos standard. Thus, we are convinced that the "harmless error" standards enunciated in Brecht and O'Neal should apply to the "trial error" before us as enunciated in those opinions and without the Ninth Circuit's modification.
For these reasons, we grant respondent's motion for leave to proceed in forma pauperis and the petition for a writ of certiorari, vacate the Ninth Circuit's determination, and remand for further proceedings consistent with this opinion.
It is so ordered.
Justice Scalia, with whom Justice Ginsburg joins as to Part I, concurring.
I do not understand the opinion, however, to address the question of what constitutes the harmlessness to which this more deferential standard is applied-and on that point the Ninth Circuit was quite correct. As we held in Sullivan v. Louisiana, 508 U.S. 275 (1993), a criminal defendant is constitutionally entitled to a jury verdict that he is guilty of the crime, and absent such a verdict the conviction must be reversed, "no matter how inescapable the findings to support that verdict might be." Id., at 279. A jury verdict that he is guilty of the crime means, of course, a verdict that he is guilty of each necessary element of the crime. United States v. Gaudin, 515 U. S. ___, ___ (1995) (slip op., at 17). Formally, at least, such a verdict did not exist here: the jury was never asked to determine that Roy had the "intent or purpose of committing, encouraging, or facilitating" his confederate's crime. People v. Beeman, 35 Cal. 3d 547, 561, 674 P. 2d 1318, 1326 (1984).
The absence of a formal verdict on this point can not be rendered harmless by the fact that, given the evidence, no reasonable jury would have found otherwise. To allow the error to be cured in that fashion would be to dispense with trial by jury. "The Sixth Amendment requires more than appellate speculation about a hypothetical jury's action, or else directed verdicts for the State would be sustainable on appeal; it requires an actual jury finding of guilty." Sullivan, supra, at 280. The error in the present case can be harmless only if the jury verdict on other points effectively embraces this one or if it is impossible, upon the evidence, to have found what the verdict did find without finding this point as well. See Carella v. California, 491 U.S. 263, 271 (1989) (Scalia, J., concurring). I concur in the remand so that the Ninth Circuit may determine whether there is "grave doubt" that this is so, rather than (what it did) determine whether it is impossible to "be certain" that this is so, 81 F. 3d 863, 867 (1996). Elsewhere in its opinion, the Ninth Circuit purported to be applying the O'Neal standard, stating that "[w]hen the reviewing court is unable to conclude the jury necessarily found an element that was omitted from the instructions," it "can only be `in grave doubt as to the harmlessness of the error,' " 81 F. 3d, at 868 (quoting O'Neal v. McAninch, 513 U. S. ___, ___ (1995) (slip op., at 3)). That seems to me to impart to the determination a black-and-white character which it does not possess, any more than other determinations possess it. It can be "the better view," but far from "certain," that, given the facts in the record, no juror could find x without also finding y. What O'Neal means is that, when the point is arguable, the State's determination of harmless error must be sustained.