OPINION

REINHARDT, Circuit Judge:

We must once again examine the question whether a state court decision denying a petitioner's ineffective assistance of counsel claim was contrary to or involved an unreasonable application of clearly established federal law as determined by the Supreme Court — this time in light of the Supreme Court's recent decision in Williams v. Taylor, 120 S.Ct. 1495 (2000). Petitioner, Tuan Tran, appeals the district court's denial of his petition for a writ of habeas corpus, claiming that he received ineffective assistance of counsel because his lawyer did not pursue motions to suppress evidence based on a lack of probable cause to arrest him and the use of an impermissibly suggestive lineup. Tran also contends that the incourt identification of his co-defendant during his trial violated his due process rights.

FACTS

Petitioner Tuan Van Tran (hereinafter "Petitioner" or "Tran") was arrested on February 11, 1989, by San Diego police officers who were investigating a series of residential robberies, including a robbery-murder. That day, Petitioner had driven co-defendant Tho Tran (not a relative) in his car to Tho's brother's apartment. At the time, the police were executing a search warrant at the apartment, as Tho was a suspect in the robberies. Shortly thereafter, investigating officer Detective Larmour noticed that Petitioner was sitting in a car parked behind the apartment and approached him. Petitioner identified himself as "Tuan Tran" and produced his driver's license. Larmour then took him to the apartment and arrested him.

Larmour testified, both at a preliminary hearing and at trial, that he based his arrest of Petitioner on several facts known to him at the time of the arrest. First, Petitioner's name, Tuan, was the same as the name that a robbery victim had heard used by one of the robbers during an attack. Second, according to Larmour, Petitioner was dressed entirely in black, as were the suspects in the robberies. Third, he fit the description of the suspects, as he was 20 to 25 years old and between 5'5" and 5'9", and he spoke in Vietnamese. Fourth, Larmour testified that Tho Tran and the Petitioner told him inconsistent stories about how they had arrived at the scene.

After his arrest, Petitioner consented to a search of his automobile where a knife was found. He was then taken to a hospital for a physical examination and collection of evidence. While at the hospital, police seized Tran's clothing and his gold ring, which was later identified as one stolen from a victim's mother. Tran was also fingerprinted at the station that evening. Police tested his fingerprints and found that they matched some of those found at the scene of the robberymurder that took place on February 2. While in custody, Petitioner was placed in a live lineup. Five witnesses identified him through that procedure as one of the robbers.

Roughly three weeks after Petitioner was arraigned, defense counsel filed a motion to suppress evidence arising from the arrest, including Petitioner's fingerprints and the lineup identification evidence, on the ground that he had been arrested without probable cause. This motion was later withdrawn by defense counsel for unexplained "tactical reasons." In addition, defense counsel prepared, but never filed, a motion to suppress the out-of-court and in-court identifications of Tran on the ground that they were too suggestive to comport with Fifth Amendment standards.

At trial, the evidence against Petitioner consisted of the following: He was named Tuan, and a victim during one of the robberies had heard a robber referred to as Tuan. He fit the description of one of the robbery suspects. He wore a gold ring that was later identified as one stolen at the first robbery, on January 18, 1989. His fingerprints matched prints found at the scene of the second robbery (on February 2), and five witnesses identified him at the lineup. Some identified him at trial as well.

Petitioner denied participating in any of the robberies or the murder. At trial, he presented a defense of mistaken identity and alibis. During the trial, the prosecution brought in codefendant Tho Tran for Detective Larmour to identify. Tho appeared in civilian clothes, without handcuffs. He did not testify. This identification is the subject of Petitioner's due process claim.

A San Diego Superior Court jury found Petitioner guilty of one count of first degree murder and three counts of residential robbery. On appeal, Petitioner argued that he received ineffective assistance of counsel because counsel failed to pursue the motions to suppress, and that the identification of co-defendant Tho in court violated due process. The California Court of Appeal, Fourth Appellate District, affirmed the conviction, holding that Tran suffered no prejudice from counsel's failure to pursue the motions to suppress, and that the due process violation was harmless error. People v. Tran, Case No. D011037, slip op. at 17-19 (Cal.Ct.App. 4th Dist. Div. 1 Jan. 23, 1991). The California Supreme Court denied Tran's appeal without opinion. Petitioner subsequently filed habeas petitions in state court which were denied. After two earlier federal habeas petitions were dismissed for failure to exhaust state remedies, the petition before us was filed, on December 18, 1996. A magistrate recommended denying the petition, and the district court adopted the reasoning set forth in his recommendation, along with some further comments of its own.

ANALYSIS

Because Tran's petition was filed on December 18, 1996, we review it under the provisions of the Anti-Terrorism and Effective Death Penalty Act (AEDPA). Lindh v. Murphy, 521 U.S. 320 (1997). Tran argues that AEDPA should not govern his petition because he first sought habeas review in 1993, prior to the date of AEDPA's enactment. Both that first petition and another petition that he filed in January 1995 were dismissed for failure to exhaust state remedies. He asserts that his current petition "relates back" to his first petition for habeas review. However, we have held that a second petition does not relate back to a first petition where the first petition was dismissed for failure to exhaust state remedies. Henry v. Lungren, 164 F.3d 1240, 1241 (9th Cir. 1999).

Tran argues that Calderon v. United States District Court (Gordon), 107 F.3d 756 (9th Cir. 1997) requires that we look to his first petition to determine what law applies. In that case, we decided that funds to investigate unexhausted claims could be awarded to a habeas petitioner even after a petition was dismissed for failure to exhaust. However, our holding in Calderon was based on the power of a district court to authorize funds prior to the initiation of habeas litigation. See Calderon, 107 F.3d at 761-62 (citing McFarland v. Scott, 512 U.S. 849 (1994)). As the power to authorize funds and attorneys' fees constitutes a special circumstance for jurisdictional purposes, we decline to extend Calderon's holding, especially where doing so would likely create a conflict with Henry.

STANDARD OF REVIEW UNDER AEDPA

The question whether the defendant has received ineffective assistance of counsel is a mixed question of law and fact. Strickland v. Washington, 466 U.S. 668, 698 (1984); Chacon v. Wood, 36 F.3d 1459, 1465 (9th Cir. 1994). Prior to AEDPA mixed questions in habeas petitions were reviewed de novo. Moran v. Godinez, 57 F.3d 690, 699 (9th Cir. 1995); Crotts v. Smith, 73 F.3d 861, 864 (9th Cir. 1996). Under AEDPA, the standard of review we apply is governed by 28 U.S.C. § 2254(d) and the Supreme Court's recent decision in Williams v. Taylor, 120 S.Ct. 1495 (2000). Under AEDPA, we may reverse a state court's decision denying relief only if that decision is "contrary to, or involves an unreasonable application of, clearly established federal law as determined by the Supreme Court of the United States." 28 U.S.C. § 2254(d)(1).

In Williams, the Supreme Court reversed the Fourth Circuit's denial of habeas relief to a capital defendant on ineffective assistance of counsel grounds. In so doing, the Court largely resolved much of the disagreement concerning how to interpret AEDPA's provision governing federal court review of state court determinations of law. First, the Court made clear that the statute embodies no distinction between pure questions of law and mixed questions of law and fact that corresponds to its division between decisions "contrary to" federal law and decisions involving an "unreasonable application of" federal law. The holding of the Court, endorsed by six justices, was that the Virginia Supreme Court's application of ineffectiveness law, a classic mixed question, was both contrary to clearly established federal law and an unreasonable application of such law. Williams, 120 S.Ct. at 1515. Moreover, Justice O'Connor's opinion for the Court interpreting AEDPA gave an example of a state court decision that could be "contrary to" Strickland in the course of explaining what that term meant. Id. at 1519-20. Thus, both prongs of § 2254(d) apply to both questions of law and mixed questions.

However, Justice O'Connor's opinion for the Court on the meaning of § 2254(d) also made clear that "contrary to" and "involve an unreasonable application of" have distinct meanings. A state court's decision can be "contrary to" federal law either 1) if it fails to apply the correct controlling authority, or 2) if it applies the controlling authority to a case involving facts "materially indistinguishable" from those in a controlling case, but nonetheless reaches a different result. Id. at 1519-20. A state court's decision can involve an "unreasonable application" of federal law if it either 1) correctly identifies the governing rule but then applies it to a new set of facts in a way that is objectively unreasonable, or 2) extends or fails to extend a clearly established legal principle to a new context in a way that is objectively unreasonable. Id. at 1520. However, the Court recognized that these categories could overlap, and that, even for purposes of precise definition, it could sometimes be difficult to determine whether a decision, for example, unreasonably extended a rule to a new context or simply contradicted controlling authority. Id. at 1521. Similarly, it seems apparent that in some cases it may be difficult to distinguish between, on the one hand, a state court decision that is contrary to clearly established federal law by virtue of its reaching a different result upon materially indistinguishable facts, and, on the other, a particularly unreasonable application of clearly established federal law. Thus, as we have said previously, the two concepts overlap and it will be necessary in some cases to test a petitioner's allegations against both standards. Davis v. Kramer, 167 F.3d 494, 500 (9th Cir.) vacated on other grounds 120 S.Ct. 1001 (2000).

The Supreme Court did not specifically define "unreasonable" in the context of decisions involving "unreasonable applications" of federal law, but it provided some explanations to help guide us in applying the concept. First, it made clear that some erroneous applications may nonetheless be reasonable: "an unreasonable application of federal law is different from an incorrect application of federal law." Williams, 120 S.Ct. at 1522 (citing Wright v. West, 505 U.S. 277, 287 (per Thomas, J.)). At the same time, the Court rejected the interpretation, adopted in various forms by the Fourth, Fifth, Seventh, and Eleventh Circuits, that defines reasonableness on the basis of whether "reasonable jurists" could disagree about the result reached by the state court. Instead, the Court adopted an "objectively unreasonable" standard, employing the language used in decisions by the Third and Eighth Circuits. Matteo v. Sci Albion, 171 F.3d 877, 889-90 (3d Cir. 1999); Long v. Humphrey, 184 F.3d 758, 760-61 (8th Cir. 1999). We think it significant that the Third and Eighth circuits adopted that test, rather than the tests developed by other circuits, because they determined that the other circuits' tests were too deferential. The Supreme Court thus chose to adopt the interpretation of AEDPA that espoused the more robust habeas review. As the Eighth Circuit explained,

The Third Circuit has rejected [the other approaches], stating that the reasonable jurist test discourages the granting of relief by requiring federal habeas courts to hold the state court acted in a way no reasonable jurist would under the circumstances and that the outside-the-universe-ofplausible-outcomes test [of the First Circuit] excludes all but those decisions so off the mark that they approach judicial incompetence.

Long, 184 F.3d at 760 (citing Matteo, 171 F.3d at 889). Thus, we must construe the Court's opinion with its endorsement of those decisions in mind.

Although the Court stated that "unreasonable" is a "common term in the legal world and, accordingly, federal judges are familiar with its meaning," Williams, 120 S.Ct. at 1522, it pointed to no particular body of law as a suitable model. While it is true that determinations of reasonableness are common in the law, (e.g., proof beyond reasonable doubt, In re Winship, 397 U.S. 358, 364 (1970); reasonable suspicion, Terry v. Ohio, 392 U.S. 1, 21 (1968); harmless beyond a reasonable doubt, Chapman v. California, 386 U.S. 18, 23-24 (1967); reasonable speech restrictions on access to nonpublic fora, Cornelius v. NAACP Legal Defense and Education Fund, 473 U.S. 788, 808 (1985); reasonable time period for post-removal order detention, Ma v. Reno, 2000 WL 258445 (9th Cir. Apr. 10, 2000)), such determinations rarely involve a judgment as to the reasonableness of a particular piece of legal reasoning. On the contrary, federal courts have not ordinarily judged the reasonableness, rather than the correctness, of a court's applications of federal law. Still, we consider several possible precedential sources in our effort to determine how to classify erroneous applications of federal law as reasonable or unreasonable.

One possible source of helpful cases might be the law governing federal court review of agency action. The federal courts have applied a "reasonableness" test when reviewing certain legal questions under Chevron v. Natural Resources Defense Council, 467 U.S. 837 (1984). In Chevron, the Court established the rule that when Congress has left a gap in a statute and authorized a federal agency to fill that gap, the agency's interpretation is to be accorded deference as long as it is a "reasonable" interpretation of the statute. Id. at 843-44.

Despite its surface similarity to the question before us, however, the deference test in Chevron is crucially different from the test we must apply, in several ways. First, the Chevron test was developed to apply in situations in which a federal agency has unique competency to fill gaps in its own governing statute that Congress has left for it to fill. Id. at 844. When an agency does not have special competence to fill such gaps, it is not accorded Chevron deference. See, e.g., Martin v. OSHRC 499 U.S. 144, 155 (1991). Obviously, the state courts do not have special competence with respect to questions of federal law. Thus, while the state courts are coequal interpreters of the Constitution, they do not play a role analogous to that of administrative agencies that are entitled to Chevron deference. Second, where an agency's interpretation is contrary to the provisions of the statute, the federal courts are required to strike it down. Chevron, 467 U.S. at 842-43; Earth Island Institute v. Mosbacher, 929 F.2d 1449, 1452 (9th Cir. 1991) (striking down regulation as inconsistent with statute). In contrast, the Court's opinion in Williams makes clear that, in some circumstances, erroneous interpretations can be upheld on habeas because they are "reasonable." Williams, 120 S.Ct. at 1522. These are, of course, only examples of some of the distinctions inherent in our review of agency decisions and those of state courts interpreting federal law. Thus, for a number of reasons, Chevron deference sheds little light on the problem before us.

Another possible source of analogous case law might be the law relating to qualified immunity. Under that doctrine, we sometimes determine whether a reasonable law enforcement official could have believed that his conduct was lawful. See Act Up! v. Bagley, 988 F.2d 868, 871 (9th Cir. 1993). While, again, the word "reasonable" is used in both circumstances, the rationale underlying the qualified immunity test renders that standard unsuitable for application to our review of state court decisions. Qualified immunity doctrine provides immunity from suit to certain government officials in spite of their having engaged in unlawful conduct because, in some contexts, it would be wrong to hold officials liable for damages where the law they are accused of violating was not clearly established. The Supreme Court has made it clear that officials whose jobs require them to perform discretionary functions must be able to fulfill their duties vigorously, and that this requires that they not be held liable for unlawful conduct where the law they violated was uncertain. Butz v. Economou, 438 U.S. 478, 507-07 (1978). In contrast, our review of state court decisions under AEDPA is not constrained because the state courts' functions are somehow more discretionary than ours, or because they must be more vigorous in the discharge of their duties. Rather, the Court has stated that deferential review in habeas cases is premised on the fact that the state courts, as part of a co-equal judiciary, are competent interpreters of federal law deserving of our full respect. Williams, 120 S.Ct. at 1518 (citing Miller v. Fenton, 474 U.S. 104, 112 (1985)). The Congressional policy limiting our authority to reverse state court judgments is motivated, therefore, by dictates of comity, rather than by a sense that the state courts, like law enforcement officials, should not be held responsible for failing to follow carefully the requirements of federal law.

Instead of relying on these faux analogues for guidance in determining what econstitutes an "unreasonable application" of federal law, we look to the doctrine of "clear error" as the source for the most helpful body of precedent. The clear error doctrine guides our review for two reasons. First, unlike the other doctrines described above, we have used "clear error" to examine and sometimes to uphold erroneous decisions of other courts on questions of law. Second, we have developed the "clear error" standard, as applied to questions of law (e.g. in mandamus cases), as part of a more general test designed to preserve "the mutual respect that is an indispensable element of the relationship between federal trial and appellate courts." In re Cement Antitrust Litigation, 688 F.2d 1297, 1301 (9th Cir. 1982) (applying clear error of law standard to mandamus petition). The "mutual respect" that informs the use of the clear error standard is highly analogous to the comity concerns at issue in habeas cases. We stress that this does not mean that reversals in habeas cases should be as rare as the granting of mandamus petitions. The Supreme Court rejected extremely deferential approaches to habeas review in favor of an approach that did not aim to "discourage the granting of relief." Long, 184 F.3d at 760. Moreover, clear error is only one of five factors involved in making mandamus decisions, and we have long held that mandamus is a "drastic" and "extraordinary" remedy. Washington Public Utilities, 843 F.2d at 323. Nevertheless, we believe the clear error doctrine — as developed not only in the mandamus context, but in other legal contexts as well, such as law of the case — in which federal courts have determined whether legal conclusions were clearly erroneous, most nearly reflects the kind of respect for other courts' judgments that Congress and the Court felt was required here.

Although the definition of clear error we have employed in differing contexts varies to some extent, it generally allows for reversal only where the court of appeals is left with a "definite and firm conviction" that an error has been committed. See, e.g., Washington Public Utilities Group v. United States District Court (Puchall), 843 F.2d 319, 325 (9th Cir. 1988) (applying "firm conviction" test to determine if legal judgment was clearly erroneous for purpose of mandamus action); In re Cement Antitrust Litigation, 688 F.2d 1297, 1301 (9th Cir. 1982) (same); Topanga Press v. City of Los Angeles, 989 F.2d 1524, 1527 (9th Cir. 1993) (reviewing grant of preliminary injunction under firm conviction test) United States v. Nelson, 137 F.3d 1094, 1106 (9th Cir. 1998) (analyzing motion for mistrial under firm conviction test where result turned on question of law (sufficiency of the evidence)); Peterson v. Highland Music, 140 F.3d 1313, 1323 (9th Cir. 1998) (applying firm conviction test to review of contempt finding, where finding requires determination that contempt was proven by clear and convincing evidence); United States v. Barbosa, 906 F.2d 1366, 1370 (9th Cir. 1990) (holding that firm conviction test applies to review of district court's application of Sentencing Guidelines to the facts); see also Sawyer v. Whitley, 505 U.S. 333, 372 (1992) (Stevens, J., concurring in part and dissenting in part) (advocating use of firm conviction standard for considering whether death sentence should have been imposed when reviewing defaulted claims on habeas).

We believe that the clear error standard occupies the middle ground that the Williams Court marked out when it rejected the arguments of those who contended that an independent determination of prejudicial error by a federal court was sufficient and of those who argued for the overly deferential "reasonable jurists" standard. Therefore, we hold that under AEDPA we must reverse a state court's decision as involving an "unreasonable application" of clearly established federal law when our independent review of the legal question does not merely allow us ultimately to conclude that the petitioner has the better of two reasonable legal arguments, but rather leaves us with a "firm conviction" that one answer, the one rejected by the court, was correct and the other, the application of the federal law that the court adopted, was erroneous — in other words that clear error occurred.

With respect to the appropriate source of law, Justice O'Connor's opinion for the Court establishes that the only definitive source of clearly established federal law under AEDPA is the holdings (as opposed to the dicta) of the Supreme Court. Williams, 120 S.Ct. at 1523. This is consistent with our interpretation of the "clearly established federal law" requirement as set forth in our pre-Williams jurisprudence. As we explained in Duhaime v. Ducharme, 200 F.3d 597, 600-01 (9th Cir. 1999), there is still a role for circuit law in habeas cases: we still look to our own law for its persuasive authority in applying Supreme Court law; however, only the Supreme Court's holdings are binding on the state courts and only those holdings need be "reasonably" applied.

[The new provision] does not mean that Ninth Circuit caselaw is never relevant to a habeas case after AEDPA. Our cases may be persuasive authority for purposes of determining whether a particular state court decision is an "unreasonable application" of Supreme Court law, and also may help us determine what law is "clearly established." See MacFarlane v. Walter, 179 F.3d 1131, 1139 (9th Cir. 1999) (looking to Ninth Circuit caselaw to confirm that Supreme Court case clearly establishes a legal rule); citing [sic] O'Brien v. Dubois, 145 F.3d 16, 25 (1st Cir. 1998) (holding that "to the extent that inferior federal courts have decided factually similar cases, reference to those decisions is appropriate in assessing the reasonableness vel non of the state court's treatment of the contested issue").

Duhaime, 200 F.3d at 600-01. See also Dows v. Wood, 2000 WL 489599, *4 (9th Cir. Apr. 27, 2000) (adopting reasoning from on-point Ninth Circuit case to help decide habeas petition under AEDPA). In addition, we have held that the Supreme Court need not have addressed the identical factual circumstance at issue in a case in order for it to have created "clearly established" law governing that case. See MacFarlane v. Walter, 179 F.3d 1131, 1139 (9th Cir. 1999). However, we may not, of course, reverse a state court's decision simply because it is inconsistent with a rule established by a Ninth Circuit case.

Finally, we must address a methodological concern with respect to our review of habeas petitions under AEDPA. Respondent's brief argues that we should proceed directly to the question whether the state court's decision was reversible under AEDPA, and implicitly suggests that we need not first consider whether the state court's decision was erroneous, as long as we find that it was not contrary to and did not involve an unreasonable application of clearly established federal law as determined by the Supreme Court. We disagree. Although the Court did not speak to this issue in Williams, in another recent case the Court employed the approach we adopt here. In Weeks v. Angelone, 120 S.Ct. 727, 729, 734 (2000), the Court first addressed the question whether the state court decision was erroneous and then, on the basis of its answer, concluded that AEDPA barred relief, rather than asking initially whether the state court decision was unreasonable under that statute. The Court also endorsed this approach in another context, stating that federal courts should first say what the law is, even when that inquiry alone does not determine whether relief should be granted in a particular case. See Wilson v. Layne, 526 U.S. 603, 609 (1999) (holding that, in qualified immunity cases, courts should address whether law was violated prior to determining whether it was clearly established); County of Sacramento v. Lewis, 523 U.S. 833, 841 n. 5 (1998) (same); see also Wilkinson v. Russell, 182 F.3d 89, 112-13 (2d Cir. 1999) (Calabresi, J., concurring) (discussing benefits of resolving status of law in qualified immunity cases). Requiring federal courts to first determine whether the state court's decision was erroneous, prior to considering whether it was contrary to or involved an unreasonable application of controlling law under AEDPA, promotes clarity in our own constitutional jurisprudence and also provides guidance for state courts, which can look to our decisions for their persuasive value. Duhaime, 200 F.3d at 600-01. Such a rule also respects our duty, as Article III judges, to say "what the law is." Williams, 120 S.Ct. at 1505 (Stevens, J., concurring) (citing Marbury v. Madison, 1 Cranch 137, 177 (1803)). Accordingly, we hold that, when analyzing a claim that there has been an unreasonable application of federal law, we must first consider whether the state court erred; only after we have made that determination may we then consider whether any error involved an unreasonable application of controlling law within the meaning of § 2254(d).

LEGAL ISSUES

Tran raises three alleged reversible errors by the state court. With respect to two of his claims, the in-court identification of his co-defendant and the lineup evidence, we hold that no error occurred. With respect to the third, probable cause for his arrest, we hold that the state court erred, but that the error was not "contrary to" and did not "involve an unreasonable application of" clearly established federal law as determined by the Supreme Court.

1. Due Process

Tran contends that his due process rights were violated when, during the trial, co-defendant Tho Tran was brought into the courtroom and identified by Detective Larmour. Tran argues that the identification evidence was irrelevant, as he did not contest that Tho was the person arrested with him. The state court agreed, but found the error harmless.

The state court did not err in holding that the error was harmless. Tran's counsel had already referred to police surveillance of Tho and his subsequent arrest and identification, and Detective Larmour had testified that the police were monitoring Tho's house because he was a suspect. Petitioner wanted the jury to believe that he was not involved in the robberies even though he gave Tho a ride. Therefore, that the jury knew that Tho was in police custody could not have prejudiced the defense's case. As the state court did not err in concluding that any error was harmless, it follows that its decision is not reversible under AEDPA.

2. Ineffective Assistance of Counsel

Both of Tran's remaining claims involve ineffective assistance of counsel. To receive relief under Strickland v. Washington, 466 U.S. 668, 698 (1984), Tran must show that his counsel's performance fell outside "the wide range of professionally competent assistance" 466 U.S. at 690, and that "there is a reasonable probability that, but for counsel's unprofessional errors, the result of the proceeding would have been different." Id. at 694.

Tran asserts that his counsel was ineffective because he failed to make motions to suppress (1) identification evidence on the basis of a suggestive lineup and (2) evidence obtained as a result of his arrest because the arrest occurred without "probable cause." The Supreme Court has held that counsel's failure to file a motion to suppress evidence can provide the basis for a claim of ineffectiveness. Kimmelman v. Morrison, 477 U.S. 365 (1986). Kimmelman held that in order to show prejudice when a suppression issue provides the basis for an ineffectiveness claim, the petitioner must show that he would have prevailed on the suppression motion, and that there is a reasonable probability that the successful motion would have affected the outcome. Id. at 375.

The state court ruled for the state on prejudice grounds, and therefore did not consider the question whether Tran's counsel's performance was deficient. This practice is specifically authorized in Strickland. 477 U.S. at 375. We, too, will begin our analysis by examining the question whether Tran would have prevailed on the motions had counsel pursued them.

A. Lineup Identifications

When analyzing lineup claims, a reviewing court must determine if the procedures in question were "unnecessarily suggestive." Neil v. Biggers, 409 U.S. 188, 198 (1972). Even if the procedures were suggestive, the resulting evidence is only excludable if it is not reliable. Id. at 199-200. Here, the state court did not err in concluding that the lineup was not unlawful. The court held that, although some of the fillers in the lineup were "arguably somewhat heavier, thinner, taller or shorter, unkempt, or older than Tran" there was no absolute requirement that the other persons in the lineup be "nearly identical" to Tran. It concluded that "reviewing the exhibits and exercising our independent judgment, Tran does not so obviously stand out as to create a substantial likelihood of misidentification."

We have reviewed the photograph of the lineup and agree with the state court's conclusion. Moreover, this circuit has rejected similar claims involving lineups where the defendant looked fairly similar to others in the lineup. United States v. Barrett, 703 F.2d 1076, 1084-85 (9th Cir. 1983); United States v. Collins, 559 F.2d 561, 563 (9th Cir. 1977). As there was no prejudice from the failure to pursue a motion to suppress the lineup identifications, we need not consider whether counsel's failure constituted deficient performance.

B. Probable Cause

The state appellate court's probable cause ruling presents a much closer question. Although we believe that the ruling was erroneous, it was not "contrary to" clearly established federal law; more important, the error did not involve an unreasonable application of federal law. Put differently, there was error, but not clear error.

The controlling law with respect to probable cause has been clearly established by the Supreme Court. "Probable cause exists where `the facts and circumstances within [the officers'] knowledge and of which they had reasonably trustworthy information (are) sufficient in themselves to warrant a man of reasonable caution in the belief that' an offense has been or is being committed." Brinegar v. United States, 338 U.S. 160, 175-76 (1949) (internal quotation omitted); Wong Sun v. United States, 371 U.S. 471, 479 (1963); Dunaway v. New York, 442 U.S. 200, 208 n. 9 (1979). The Supreme Court has applied this test to arrests, holding that probable cause requires that the officers have similarly sufficient information to believe that the individual to be arrested has committed the crime in question. Dunaway, 442 U.S. at 208 n. 9. The state court cited this standard, and clearly intended to apply it. As none of the Supreme Court's cases involve facts materially indistinguishable from those at issue here, the state court's decision was not contrary to clearly established law. The ultimate question before us, therefore, is whether the state appellate court unreasonably applied the controlling federal law.

As our above analysis illustrates, the clearly established federal law, the application of which must be determined to be reasonable or unreasonable, is ordinarily to be found at the level of generality of a general rule. Where there is an overall multi-factor test, for example, the clearly established federal law consists of the multi-factor test itself, not of any particular applications of the test. Thus, while the probable cause test itself constitutes clearly established federal law, the past applications of the test come into play only at a later stage in our analysis — when we decide whether the application before us is "contrary to" or an unreasonable application of the rule in question. If an application of a rule in a case before us conflicts with either the rule itself or a past application of the rule by the Supreme Court, then the case will involve a "contrary to" question. Otherwise, it will involve a reasonable application issue.

Of course, we cannot set forth here a precise mechanism for determining the appropriate level of generality with respect to every type of legal question. Suffice it to say that we must refrain from applying too narrow or too specific a standard when determining the substance of the law the application of which we must assess. The adoption of too limited a scope for the definition of "clearly established law" would collapse the various parts of the analysis we must perform and prevent a proper application of the Williams mandate.

In determining that the arresting officer had probable cause, the state appellate court reported the following facts in its opinion: (1) Tran and his co-defendant gave inconsistent stories about how they arrived at the apartment that the police were searching, (2) Tran was in a car with an identified robbery suspect on the same day a robbery occurred, and admitted that he and the suspect traveled to the apartment together, (3) Tran's first name matched the name of a robber heard by a victim of one of the robberies, (4) Tran fit the general description of one of the robbers. The state court concluded that these facts established probable cause, given the totality of the circumstances. There is one problem with the facts as reported, however. There is no evidence in the record that the arresting officer knew that the suspect, Tho, was involved in a robbery that had occurred that day (or that any robbery at all had occurred that day). Accordingly, we must disregard that specific aspect of the state court's rationale. However, we do take into account related facts that appear in both the opinion and the record: namely, that the officer knew that there had been a series of robberies that had occurred in the period prior to the arrest, that Tho was a suspect in those robberies, and that the last robbery of which the officer had knowledge occurred nine days before petitioner's arrest.

After carefully considering the factors mentioned above, we conclude that the state court's decision was in error, but that the error was not objectively unreasonable. Petitioner fit the physical description of one of the suspects — he appeared to be between 18 and 20 years of age, was of slight build, Vietnamese, clean shaven, and approximately 5' 5". While we have previously found a description of "a young, thin man, not too tall" and "a young Mexican kid" in United States v. Ricardo D., 912 F.2d 337, 338 (9th Cir. 1990), to be insufficient to warrant a probable cause finding, in that case the description was both general and constituted the primary basis for the arrest. Here, in contrast, the description was more specific. The police knew a specific age range, a specific height range, a description of the suspect's build and (lack of) facial hair, and a definitive racial identification. Moreover, this was not the only information on which the officers relied.

The officers also knew that one of the suspects was named "Tuan." This factor would be entitled to substantial weight were the petitioner's name unusual. However, evidence in the record shows that the robberies in question were occurring within the large Vietnamese community in San Diego, and that Tuan is an extremely common name in that community. For example, in this case, the Petitioner, one of the witnesses, and another suspect were all named Tuan. Nonetheless, the identity of the names is of some value, and the court was entitled to give it some weight.

The state court's reasons, as we modify them, see n. 20 supra, also included the fact that Petitioner was in the company of a suspected robber who had been engaged in a continuous series of robberies. Although the mere association of someone with a criminal suspect does not give rise to probable cause, Ybarra v. Illinois, 444 U.S. 85, 91 (1979), here the association was to some extent connected in time to suspicious activity, both because Tran was driving Tho to a location that was the subject of a search warrant in connection with the robberies and because Tho was a suspect in an ongoing series of robberies.

Finally, Petitioner and co-defendant Tho Tran gave inconsistent stories regarding how they arrived at the apartment. While Petitioner claimed that he had driven Tho there, Tho claimed that he had walked. The fact that Petitioner was telling the truth does not negate this factor, because at the time the arrest was made Tho's lie gave the police reason to believe that one of the two was lying, and therefore suggested that there was something to cover up.

The state court ruled that, considering the totality of the relevant factors, there was probable cause for Tran's arrest. See Illinois v. Gates, 462 U.S. 213, 238 (1983) (adopting totality of circumstances test for probable cause determinations). Were we reviewing this case de novo, we would hold that the state court's decision was erroneous; the facts adduced here were not enough to cause a person of reasonable caution to believe that Tran had participated in the robberies. The temporal connection between the robberies and Tran's association with Tho was insufficiently close to constitute relevant association under Ybarra. See United States v. Hillison, 733 F.2d 692, 697 (9th Cir. 1983), and United States v. Hoyos, 892 F.2d 1387, 1393 (9th Cir. 1988). In the absence of such an association, the other factors are simply insufficiently substantial to warrant a probable cause finding. Cases upholding arrests for probable cause typically involve considerably more evidence than this. See, e.g., United States v. Meza-Corrales, 183 F.3d 1116, 1124 (9th Cir. 1999); United States v. Valencia, 24 F.3d 1106 (9th Cir. 1994); United States v. Hoyos, 892 F.2d 1387 (9th Cir. 1989). Nevertheless, after reviewing the record and the relevant cases, we cannot say, applying a firm conviction standard, that the court's decision was clearly erroneous. The issue is simply too close. Accordingly, we conclude that the state court's decision did not involve an unreasonable application of clearly established federal law.

CONCLUSION

Having found that the state court's decision was not contrary to, and did not involve an unreasonable application of, clearly established federal law, we affirm the district court's denial of Tran's habeas petition.

AFFIRMED.