Ronnie Earl Baylor’s trial counsel failed to follow-up on a criminalist’s report that concluded that a semen sample taken from one of the victims he was convicted of sexually assaulting might not have come from Baylor. Baylor’s petition for writ of habeas corpus filed in the California Supreme Court, which raised ineffective assistance of counsel, was denied without citation to authority; his federal petition pursuant to 28 U.S.C. § 2254 was granted following an evidentiary hearing. The state has timely appealed, challenging the writ under the “old law,” and arguing in addition that the new provisions of § 2254 (enacted April 24, 1996, after the district court’s decision was rendered and opening briefs were filed) in the Antiterrorism and Effective Death Penalty Act of 1996, Pub.L. 104-132, 110 Stat. 1214, (Antiterrorism Act), apply to this appeal and also require reversal.

Because we affirm the district court’s writ of habeas corpus, we must decide whether § 2254(d), upon which the state relies, re quires a different result. Whether or not the new Act will apply to other eases on appeal, Strickland v. Washington, 466 U.S. 668 , 104 S.Ct. 2052 , 80 L.Ed.2d 674 (1984), has long been clearly established federal law determined by the Supreme Court of the United States. We therefore conclude that our analysis in this ease is not affected by § 2254(d).

We have jurisdiction, 28 U.S.C. §§ 1291 , 2253, and affirm.

Baylor was charged with forcible rape, forcible sodomy, assault with intent to commit rape and related burglary and robbery of two victims occurring in December 1986 and January 1987. Several months before trial, David Stockwell, a criminalist in the Sheriff’s Crime Laboratory, issued a report in which he compared a blood and saliva sample from Baylor with seminal fluid from vaginal swabs obtained from the first victim, and with semen stains from items of clothing from the second victim. Stockwell concluded that the results would tend to eliminate Baylor as the semen donor because Baylor is a “secretor” and the semen donor wasn’t, but that there is a possibility that Baylor mimics a nonse-cretor and that this theory could not be tested without a liquid semen sample from Baylor. Baylor’s counsel knew about the report before trial but neither followed-up with Stockwell nor had Baylor tested by another expert. While counsel tried to subpoena Stockwell during trial, it turned out that Stockwell was on vacation; without a foundation, the report was, of course, inadmissible.

At trial, no witness identified Baylor and there was no physical evidence linking him to the crimes. However, Baylor had confessed, in some detail, on tape, to detective Roderick, who testified; Baylor also testified, offering an alibi (which his wife corroborated for the first victim) and recanting his confession, saying that he confessed only because Roderick threatened him and fed him the details.

The jury convicted Baylor on all counts, and he was sentenced to a term of 16 years in custody. After his sentence was affirmed on direct appeal, Baylor filed a petition for writ of habeas corpus with the California Supreme Court June 19, 1990, which was denied August 15, 1990 without citation to authority. Baylor then filed his petition for writ of habeas corpus in the district court, alleging that trial counsel’s failure to get the Stockwell report admitted into evidence was ineffective assistance. Although the district court initially adopted the Findings, Conclusions and Recommendations of the magistrate judge denying the petition on the ground that Baylor had not shown prejudice in the absence of evidence that he was a nonseeretor, the court granted Baylor’s motion for relief from judgment and ordered serological testing of his seminal fluid. Based on laboratory results and the opinions of both a defense expert as well as an independent expert that the semen on the vaginal swabs is genetically incompatible with originating from Baylor, the district court found that trial counsel’s performance was “completely inadequate” and that “there is at least a reasonable probability that the evidence presented would have tended to show that Petitioner could not have been the source of the semen found in [the] victim.” The court further concluded that, even in the face of the strong evidence provided by Baylor’s taped confession, “there is a reasonable probability that the evidence that would have been presented but for counsel’s ineffective assistance would have injected reasonable doubt in the jurors’ minds in regards to all of Petitioner’s convictions.”

The state argues that Baylor was not denied effective assistance for a variety of reasons that are unpersuasive to us. Among them: trial counsel tried to check on Stock-well’s availability before trial; he subpoenaed Stockwell when he found out, during trial, that the district attorney didn’t intend to call him; when he learned that Stockwell was on vacation, he tried to get the report admitted anyway, which the court declined to do; he discussed a writ of prohibition with Baylor that would have continued the trial, but Baylor didn’t “want to continue this”; and he could reasonably have decided to attack or minimize Baylor’s confession instead of pursuing Stockwell’s report, which he thought was inconclusive.

We have difficulty understanding how reasonably competent counsel would not recognize “the obvious exculpatory potential of semen evidence in a sexual assault case,” Thomas v. Goldsmith, 979 F.2d 746 , 750 n. 2 (9th Cir.1992), particularly when the criminalist’s report plainly indicates that the donor was an ABO nonsecretor whereas Baylor was an ABO type “O” secretor and that this “would thus eliminate” Baylor as the perpetrator unless a test (for the H antigen level) on a liquid semen sample showed that he mimicked a nonsecretor, in which case he would conform to the semen donor profile. Whether or not Stockwell’s report was itself conclusive, it was one test away from tilting the scale powerfully in Baylor’s direction. Even without further testing, Stoekwell’s report would have greatly decreased the odds that Baylor was the donor (as Stoekwell testified at the evidentiary hearing, there was only a 5% likelihood that a secretor would mimic a nonsecretor), and buttressed the credibility of Baylor’s alibi and recantation at trial. For these reasons, we agree with the district court that counsel did not give reasonably effective assistance. Strickland, 466 U.S. at 691 , 104 S.Ct. at 2066 (counsel “has a duty to make reasonable investigations or to make a reasonable decision that makes particular investigations unnecessary”); see, e.g., Sanders v. Ratelle, 21 F.3d 1446, 1456 (9th Cir.1994) (counsel ineffective where he knows of, yet fails to investigate, another person’s purported confession); Evans v. Lewis, 855 F.2d 631, 637 (9th Cir.1988) (counsel ineffective where documents that indicate the defendant’s history of mental problems are available yet he fails to investigate a possibility of mental impairment as a mitigating factor for sentencing purposes); see also Sims v. Livesay, 970 F.2d 1575, 1580-81 (6th Cir.1992) (counsel was ineffective where counsel knows of a potentially exculpatory gun residue report yet counsel fails to investigate adequately).

The more difficult question is whether Baylor was prejudiced, the second prong of the Strickland analysis. Baylor confessed, and his confession was corroborated by the victims’ version of the attacks. Although he testified that his confession was coerced, there is no finding that it was; also, Baylor confessed to details that the state argues would have been unknown to the victims or the detective. In particular, Baylor confessed that he removed five bullets from a gun he found in the first victim’s house and put them under the cushion of a large chair in her living room. There is evidence that these bullets were not discovered until after Baylor told Roderick where they were. Also, as the state points out, the jury credited Roderick’s testimony instead of Baylor’s. In any event, the state contends, the case against Baylor was strong even with the laboratory reports produced at the evidentia-ry hearing, because the reports were inconclusive and disputed.

Although Baylor’s confession is strong evidence that he committed both assaults, we cannot say that the district court erred in concluding that he was prejudiced. Our confidence in the outcome of the trial is shaken by the results of the test on Baylor’s semen that Stockwell’s report suggested would resolve his secretor status. The laboratory reports, and the analysis of two different experts, indicate that Baylor does not mimic a nonsecretor and therefore can be excluded as the donor of the semen found on the first victim.

The state argues that the science upon which Baylor relies isn’t reliable because, for example, serological tests performed in 1994 do not necessarily eliminate Baylor since his semen could have changed over the course of seven years, and because of the possibility that the semen recovered from the victim’s vaginal swab wasn’t Baylor’s and that Stock-well missed “H” antigens present on the original sample due to operator error. Baylor’s expert says otherwise, and after carefully considering all the evidence, the district court found that Stockwell’s reasons for concluding that the scientific evidence that would have been presented at trial may not have excluded Baylor are without scientific foundation. We cannot say that the court clearly erred.

The state also argues that even in the face of evidence arguably excluding Baylor as the semen donor, that evidence cannot overcome the force of his detailed confession. As the district court concluded, however, reasonable doubt about the rape of the first victim would necessarily have raised a reasonable doubt about the validity of his confession. Inasmuch as all of Baylor’s convictions were based on the same confession, we cannot disagree with the court’s conclusion that there is a reasonable probability that the evidence that would have been presented but for counsel’s ineffective assistance would have injected reasonable doubt in regard to all his counts. There were no witness identifications, and there was no physical evidence linking Baylor to the crime scenes. Baylor has, therefore, shown prejudice. Strickland, 466 U.S. at 694 , 104 S.Ct. at 2068 .

Ill

Prior to oral argument, the state requested an opportunity to file supplemental briefing on the issue of the retroactivity of the Antiterrorism Act. In turn, we asked both parties to brief the applicability of the Act. Having considered the supplemental briefs, we conclude that whether or not the Act applies to pending appeals, our decision would be the same. Section 2254(d), upon which the state relies in this ease, provides that a writ of habeas corpus shall not issue unless the petitioner establishes that the state court decision was contrary to clearly established federal law as determined by the United States Supreme Court. In this case, the state court decision was contrary to clearly established federal law determined by the Court in Strickland. See Ayala v. Speckard, 89 F.3d 91, 96-97 (2d Cir.1996) (adopting similar approach).

To the extent that the state suggests that the Act somehow makes the district court’s evidentiary hearing irrelevant, we disagree. It appears to argue that federal courts lack jurisdiction to review anything other than the state court judgment. Without regard to how the Act may change the way things are donfe in the future, in this case both parties agreed that the evidentiary hearing was proper and Baylor’s first chance to develop the factual basis of this claim was in the federal district court. We therefore decline to disregard what the district court did.

AFFIRMED.

. Amended section 2254(d) provides: An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim - (1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; or (2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.