concurring in the judgment.
Having dissented in Doyle v. Ohio, 426 U. S. 610, 620-635 (1976), I can readily understand why the Court might want to overrule that case. But if there is to be a rule that prohibits a prosecutor’s use of a defendant’s post-Miranda silence, it should be a clearly defined rule. Whether the trial court sustains an objection to an impermissible question, or whether the prosecutor is allowed to refer to the defendant’s silence in his or her closing arguments, are questions that are relevant to the harmless-error inquiry, or to deciding whether the error made the trial fundamentally unfair. But they play no role in deciding whether a prosecutor violated the implicit promise of Miranda — as understood in Doyle— that the defendant’s silence will not be used against him.
I, therefore, agree with the 10 Illinois judges and 12 federal judges who have concluded that the rule of the Doyle case was violated when the prosecutor called the jury’s attention to respondent’s silence. Moreover, for the reasons stated by the Court of Appeals, I think the violation was serious enough to support that court’s conclusion that the error was not harmless beyond a reasonable doubt. United States ex rel. Miller v. Greer, 789 F. 2d 438, 445-447 (CA7 1986) (en banc). Were this case here on direct appeal, therefore, I would vote to reverse the conviction.
*768Nonetheless, I concur in the Court’s judgment because I believe the question presented in the certiorari petition— whether a federal court should apply a different standard in reviewing Doyle errors in a habeas corpus action — should be answered in the affirmative. In Rose v. Lundy, 455 U. S. 509 (1982), I argued that there are at least four types of alleged constitutional errors.
“The one most frequently encountered is a claim that attaches a constitutional label to a set of facts that does not disclose a violation of any constitutional right. . . . The second class includes constitutional violations that are not of sufficient import in a particular case to justify reversal even on direct appeal, when the evidence is still fresh and a fair retrial could be promptly conducted. Chapman v. California, 386 U. S. 18, 22; Harrington v. California, 395 U. S. 250, 254. A third category includes errors that are important enough to require reversal on direct appeal but do not reveal the kind of fundamental unfairness to the accused that will support a collateral attack on a final judgment. See, e. g., Stone v. Powell, 428 U. S. 465. The fourth category includes those errors that are so fundamental that they infect the validity of the underlying judgment itself, or the integrity of the process by which that judgment was obtained.” Id., at 543-544 (dissenting opinion) (footnote omitted).
In my view, Doyle violations which cannot be deemed harmless beyond a reasonable doubt typically fall within the third of these categories. On direct review, a conviction should be reversed if a defendant can demonstrate that a Doyle error occurred at trial, and the State cannot demonstrate that it is harmless beyond a reasonable doubt. But, in typical collateral attacks, such as today’s, Doyle errors are not so fundamentally unfair that convictions must be reversed whenever the State cannot bear the heavy burden of proving that *769the error was harmless beyond a reasonable doubt. On the other hand, there may be extraordinary cases in which the Doyle error is so egregious, or is combined with other errors or incidents of prosecutorial misconduct, that the integrity of the process is called into question. In such an event, habeas corpus relief should be afforded.*
In sum, although I agree with the Court’s judgment, and the standard that it applies here, I would apply this standard only to Doyle violations being considered on collateral review. On direct appeal, a Doyle error should give rise to reversal of the conviction unless the State can prove that the error was harmless beyond a reasonable doubt.
In Rose v. Lundy, I noted that the distinction between direct and habeas review is supported by the Court’s decisions concerning retroactive application of newly recognized constitutional rights. 455 U. S., at 509, 543, n. 8 (dissenting opinion). Our recent decision in Griffith v. Kentucky, 479 U. S. 314 (1987), endorsed Justice Harlan’s view on this issue to a great extent, and thus supports this proposition. Compare Griffith, supra, with Allen v. Hardy, 478 U. S. 255 (1986).