dissenting.
The Court today properly holds that a complete denial of cross-examination designed to explore the bias of a prosecution witness violates the Confrontation Clause, whether or not the denial influenced the outcome of the trial and whether or not the witness was important to the prosecution’s case. Nevertheless, the Court remands in order to permit the state court to apply harmless-error analysis to that violation. I must respectfully dissent from the latter part of the Court’s holding. I believe that the importance of cross-examination to a criminal trial is so great that a complete denial of otherwise proper cross-examination concerning the potential bias of a prosecution witness should lead to no less than a reversal of the conviction.
*687In holding the Confrontation Clause applicable to the States, this Court referred to the right of cross-examination as “an essential and fundamental requirement for the kind of fair trial which is this country's constitutional goal.” Pointer v. Texas, 380 U. S. 400, 405 (1965). If indeed the harmless-error doctrine focuses on the fairness and accuracy of a criminal trial, see ante, at 681, it is odd that the majority so easily applies it to a type of error that calls both fairness and accuracy into question to an almost unique degree.
The centrality of cross-examination to the factfinding process makes it particularly unlikely that an appellate court can determine that a denial of cross-examination had no effect on the outcome of a trial.
“[T]he court ordinarily cannot measure whether harm has ensued to an appellant when he has been denied the opportunity to cross-examine witnesses against him, given all the risks. Had cross-examination been allowed, for example, it might have served to impeach a witness and thus to cast doubt on corroborating testimony, or it might have elicited exculpatory evidence. Only on rare occasions will an appellate court be able to find that the testimony of the witness was so tangential, or so well corroborated, or so clearly invulnerable to attack that the denial of the right to cross-examination was harmless.” R. Traynor, The Riddle of Harmless Error 68-69 (1970).
The fact that a particular witness’ testimony was corroborated cannot render harmless a denial of the right to expose his bias. Defense counsel may have valid strategic reasons for challenging one witness’ testimony aggressively while treating a corroborating witness more gently. Jurors evaluating the witnesses’ demeanor may choose to give great weight to the testimony of one witness while ignoring the similar testimony of another. In either event, denial of cross-examination concerning a witness’ bias may deprive the defense of its best opportunity to expose genuine flaws in the *688prosecution’s case — flaws that the cold record will not reveal to an appellate court.
Indeed, an appellate court attempting to apply harmless-error analysis is faced with a formidable burden. The court cannot merely satisfy itself that the jury would have reached the same result had the witness in question not appeared at all; it must be convinced beyond a reasonable doubt that the jury would have reached the same result even if cross-examination had led the jury affirmatively to. believe that the witness was lying. Moreover, the court must conclude, beyond a reasonable doubt, that no evidence exculpatory to the defendant could have emerged from a genuinely adversarial testing of the witness. I think that a court can make such a determination only in the rarest of circumstances, and a rule of per se reversal is therefore justified.
The Confrontation Clause violation in this case is especially pernicious. The jury was essentially misled, by the empty gesture of cross-examination, to believe that the defense attorney had been permitted to use all the tools at his disposal to expose weaknesses in Fleetwood’s testimony. Having survived what appeared to be counsel’s best efforts to undermine the witness’ credibility, Fleetwood’s testimony necessarily carried more weight with the jury than would the same testimony given without an apparent opportunity to cross-examine.
This analysis makes it unnecessary to strain, as does the majority, to reconcile the apparent per se rule of Davis v. Alaska, 415 U. S. 308 (1974), with the harmless-error analysis employed in Harrington v. California, 395 U. S. 250 (1969), and Schneble v. Florida, 405 U. S. 427 (1972). I would simply hold that Davis mandates reversal whenever the prosecution puts a witness on the stand but the court does not- permit the defense to cross-examine concerning relevant potential bias. I therefore dissent from the Court’s decision to permit the Delaware Supreme Court to apply *689harmless-error analysis to the Confrontation Clause violation in this case.
I also write to emphasize that this Court cannot require state courts to apply harmless-error analysis to violations of the Federal Constitution. See Connecticut v. Johnson, 460 U. S. 73, 88 (1983) (Stevens, J., concurring in judgment). While this Court has stated that federal law governs the application of harmless error to violations of the Federal Constitution, see Chapman v. California, 386 U. S. 18, 21 (1967), that cannot mean more than that state courts must reverse convictions when the Constitution so mandates. When the Constitution does not mandate a particular remedy, this Court may not “declare which of many admittedly constitutional [remedial] alternatives a State may choose.” Id., at 48 (Harlan, J., dissenting) (footnote omitted). We have never held that the Federal Constitution forbids state courts to reverse certain convictions pursuant to state substantive or procedural law, nor can I imagine what provision of the Constitution could grant us such a power. Thus the Delaware Supreme Court remains free on remand to decide that even though it applied the substantive standards of the Sixth Amendment to determine whether error occurred, its harmless-error analysis was the product of state rather than federal law.