concurring:
I concur in Part One of the majority opinion holding that the instruction constituted a Sandstrom violation and was not harmless error.
I concur in the result only of Part Two, holding that the prosecutor’s argument rendered the sentencing phase of the trial fundamentally unfair and was not harmless error under Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967). I do not agree, however, with the use of the Strickland v. Washington test in this context.
JOHNSON, Circuit Judge,specially concurring:
Although I concur in the result reached in this case, I write separately to emphasize my belief that it can be reached by a more direct route than that selected by the majority. The majority finds that the Sandstrom error cannot be harmless because the jury returned a general verdict on a defendant indicted on three theories of murder, and as to one of those theories— aiding and abetting — the evidence of intent was not “overwhelming.” This complicated inquiry into multiple theories of liability becomes necessary because of a premise adopted by the majority at the beginning of its argument: that “evidence of intent [can be] overwhelming, even where there is conflicting evidence as to whether the defendant was the killer.” 762 F.2d at 1454. This premise, which forecloses the more straightforward holding that the evidence was not “overwhelming” as to the malice murder charge, was first introduced by the majority in Davis v. Kemp, 752 F.2d 1515 (11th Cir.1985) (en banc). I rejected this approach when it was presented in Davis, and I find here, as well, that it provides an unsuitable basis on which to build the majority opinion.
The Davis majority stated that “[although some opinions talk in terms of overwhelming evidence of guilt ... the crucial inquiry relates to whether or not there is overwhelming evidence of intent.” 752 F.2d 1521 n. 10. The consequence of this understanding was that the Court was able to limit itself to the inquiry of whether the evidence that “whoever killed the victim did so with intent” was overwhelming, 752 F.2d 1521, without considering the weight of the evidence that Davis was the killer. This consequence is elevated to the status of a rule in the majority opinion in this case: a court can find “overwhelming” evidence of intent even where there is no “overwhelming” evidence that the defendant was the individual who committed the act. The flaw in this new rule is the same as that in Davis: while it is necessary to limit the court’s scrutiny to evidence of intent, because the jury’s determination of intent is the only one that could be affected by the erroneous instruction, it is also necessary to limit the court’s scrutiny to evidence of the intent of the defendant because it is his connection with the crime that is to be tested at trial.1 If a court *1470evaluates the evidence of the killer’s intent without simultaneously evaluating evidence that the defendant is the killer, it may finish by evaluating evidence wholly inapplicable to the suspect on trial. Efforts to render such evidence applicable retroactively, on the ground that the jury ultimately found that the defendant committed the act, are also inappropriate. A court reviewing for Sandstrom violations must know the strength of the evidence presented at trial, so it can assess the effect of an instruction delivered at the trial’s conclusion. If the evidence presented at trial is reinterpreted in light of conclusions reached by the jury in its verdict, the effect of the erroneous instruction on the jury’s deliberations cannot be accurately gauged. Moreover, a rule which made all inculpatory evidence retroactively applicable to the defendant whenever the jury reached a guilty verdict would render the evidence “overwhelming” in every case in which such a verdict had been reached. Davis v. Kemp, supra, 752 F.2d at 1528 n. 1. (Johnson, J., dissenting).
I would hold that, because evidence of Drake’s intent to commit the murder was not overwhelming, the Sandstrom error was not harmless beyond a reasonable doubt when measured by the rule of Chapman v. California, 386 U.S. 18, 87 S.Ct. 824, 17 L.Ed.2d 705 (1967).
. Although the Supreme Court in Francis v. Franklin, - U.S. -, 105 S.Ct. 1965, 85 L.Ed.2d 344 (1985) concluded that it was evidence of intent rather than evidence of guilt which must be "overwhelming," the Court neither considered nor affirmed the further conclusion that evidence of intent may be overwhelming where evidence that the defendant was the *1470killer is not. In fact one standard applied by the Court in its harmless error analysis (requiring "evidence so dispositive of intent that a reviewing court can say beyond a reasonable doubt that the jury would have found it unnecessary to rely on the presumption” (citing Connecticut v. Johnson, 460 U.S. at 97 n. 5, 103 S.Ct. at 973 n. 5 (Powell, J., dissenting)), - U.S. -, 105 S.Ct. at 1977, provides additional support for the proposition that none but the most comprehensive and compelling evidence of intent would be sufficient to render a Sandstrom error harmless. While the question remains to be addressed by the Court, it seems unlikely that the evidence would be construed to meet this elevated standard where it failed overwhelmingly to establish the identity of the killer.