FOR PUBLICATION
 UNITED STATES COURT OF APPEALS
      FOR THE NINTH CIRCUIT

HENRY EARL DUNCAN,                        No. 05-99010
            Petitioner-Appellant,
              v.                            D.C. No.
                                         CV-92-01403-AHS
STEVEN W. ORNOSKI,
                                            OPINION
            Respondent-Appellee.
                                    
       Appeal from the United States District Court
           for the Central District of California
      Alicemarie H. Stotler, District Judge, Presiding

                 Argued October 26, 2007
                 Submitted March 6, 2008
                   Pasadena, California

                   Filed June 24, 2008

     Before: Stephen Reinhardt, Ronald M. Gould, and
              Richard A. Paez, Circuit Judges.

               Opinion by Judge Reinhardt




                           7307
7310                 DUNCAN v. ORNOSKI


                        COUNSEL

Mitchell Zimmerman, Mountain View, California, Richard C.
Neuhoff, New Britain, Connecticut, Kathryn J. Fritz and
Albert Sieber, San Francisco, California, for the petitioner-
appellant.
                        DUNCAN v. ORNOSKI                   7311
Edmund G. Brown, Jr., Attorney General of the State of Cali-
fornia, Dane R. Gillette, Chief Assistant Attorney General,
Pamela C. Hamanaka, Senior Assistant Attorney General,
Keith H. Borjon, Supervising Deputy Attorney General, Scott
A. Taryle, Supervising Deputy Attorney General, Los Ange-
les, California, for the respondent-appellee.


                            OPINION

REINHARDT, Circuit Judge:

   Once again, we consider whether a capital defendant’s
appointed lawyer’s performance was so deficient and prejudi-
cial that it violated his Sixth Amendment right to counsel.
Appellant Henry Earl Duncan was convicted of robbery and
first-degree murder on March 3, 1986. The jury found the spe-
cial circumstance allegation to be true and, after a brief pen-
alty phase hearing, sentenced Duncan to death. The California
Supreme Court affirmed the judgment on direct appeal and
subsequently denied Duncan’s petition for writ of habeas cor-
pus on the merits. Duncan filed a federal habeas petition in
the Central District of California. The district court denied
most of his claims and then held a four-day evidentiary hear-
ing, after which it rejected the rest. Duncan appeals.1

   We conclude that Duncan’s lawyer’s performance was
deficient during the guilt phase of his trial because he failed
to investigate and present evidence that the blood samples
from the crime scene that did not belong to the victim also did
not belong to Duncan. This evidence would have tended to
establish that Duncan had an accomplice who was in the mur-
der room on the night of the murder, shed blood, and used the
first aid kit on the wall to treat his wounds. Indeed, the evi-
dence would have been sufficient to support an inference that
  1
   We have jurisdiction under 28 U.S.C. §§ 1291 and 2253.
7312                      DUNCAN v. ORNOSKI
it was the accomplice, not Duncan, who killed the victim.
Nevertheless, evidence with respect to Duncan’s presence at
the crime scene on the night of the murder, including his shoe
prints, fingerprint, and palm prints in the money room, is suf-
ficient to show that Duncan participated in the robbery and
thus to sustain Duncan’s conviction for felony murder.
Accordingly, we hold that Duncan’s lawyer’s deficient perfor-
mance did not prejudice him with respect to his conviction.
However, counsel’s failure to investigate and present the
potentially exculpatory serological evidence did prejudice
Duncan with respect to the jury’s special circumstance find-
ing, which, under California law at the time of his trial,
required proof beyond a reasonable doubt that he intentionally
killed the victim or, if not, that he intended that she be killed.
Because the serological evidence raises doubts as to whether
Duncan was the actual killer, and the evidence in the record
does not establish beyond a reasonable doubt that Duncan
intended that the victim be killed, we conclude that counsel’s
ineffective performance was prejudicial and thus constituted
a Sixth Amendment violation. Accordingly, we reverse the
judgment in part and remand with instructions to grant the
petition as to the jury’s special circumstance finding and to
vacate the sentence.2

I.       Factual and Procedural Background

     A.    The Crime

   At the time of the murder, Duncan worked as a cashier at
the International Host Restaurant in the Los Angeles Interna-
     2
    Duncan also asserts that he received ineffective assistance of counsel
in the penalty phase. Although his claim is likely meritorious, we need not
reach the question because our holding necessitates vacating the sentence.
We address Duncan’s remaining claims that relate to the guilt phase of his
trial in a separate memorandum disposition filed concurrently with this
opinion. Because we find no merit to those claims, we affirm the district
court’s denial of habeas relief as to his conviction for first-degree felony
murder.
                      DUNCAN v. ORNOSKI                    7313
tional Airport. The murder victim, Josephine Eileen DeBaun,
was his supervisor. As part of her closing duties each night,
DeBaun balanced the day’s receipts and deposited them in a
safe located in a small caged area in the restaurant’s back
office. This caged area was commonly called “the money
room.” DeBaun was murdered in the money room on the
night of November 13, 1984.

   On that evening, Duncan completed his regular shift and
clocked out around 11:00 p.m. An airport custodian saw Dun-
can not far from the restaurant around 11:30 p.m., and Dun-
can told her that he was waiting for someone. DeBaun was
last seen alive around 11:50 p.m. by another worker before he
left the restaurant.

   DeBaun’s body was discovered on the floor of the money
room the next morning. She had sustained multiple stab
wounds and blood covered the tiny room. A broken knife han-
dle was found on the floor next to her body. The supervisor’s
“floating fund bank”—a locked metal drawer used to store
cash—was found open with $2100 missing and a ring of keys,
including the VM-19 key used to open the bank, dangling
from the lock. The ring of keys, commonly called the “boss
key ring,” was usually kept inside a padlocked key box that
was mounted on the wall. Also kept in the box, but hidden,
was an additional copy of the VM-19 key, which opened the
floating fund bank. Police discovered the padlock to the key
box underneath a bloodied Handiwipe rag that was sitting on
a table near the body just below an open first aid kit that hung
on the wall. The contents of the open first aid kit were dis-
turbed, which suggested to police investigators that the assail-
ant was injured during the attack and took some first aid
supplies to treat his wounds. There were also a number of
defensive wounds on the victim’s hands and fingers, which
indicated that there had been a struggle before the murder.

  Police investigators found various bloody palm prints, a
bloody shoe print, and a bloody fingerprint in the money
7314                      DUNCAN v. ORNOSKI
room. Duncan’s fingerprints and palm prints, along with those
of fifteen other individuals, were taken shortly after the crime.
A fingerprint expert at the Los Angeles Police Department
determined that Duncan’s prints were not a match for the
prints found at the crime scene. Police also took various blood
samples from the crime scene and performed serological tests
that showed that some of the blood did not belong to the vic-
tim.

   A second robbery occurred at the restaurant three months
later, in which $1770 was stolen from the supervisor’s float-
ing fund bank, which was opened with the duplicate VM-19
key. Three hundred dollars was also missing from Duncan’s
own cashier bank. Duncan was arrested for the second rob-
bery. His prints were taken again and with that additional evi-
dence, the bloody prints at DeBaun’s murder scene were
determined to be his.3 The bloody shoe print was then found
to be “similar in class characteristics” to a pair of shoes found
in Duncan’s house, and the duplicate VM-19 key was found
in Duncan’s car.

   Duncan was charged with robbery and murder, and sepa-
rately with grand theft. He subsequently pled guilty to the
grand theft charge in connection with the second incident, and
was tried on the robbery and murder counts in connection
with the first. The Information alleged that Duncan personally
killed DeBaun with a dangerous weapon and the special cir-
cumstance that he did so during the course of a robbery.

  B.    The Trial

  John Cheroske, who had originally been retained by Dun-
can’s mother to represent Duncan at the preliminary hearing,
  3
    According to the fingerprint expert, there was a mistake in the way the
initial left index print had been taken the first time, and he was not able
to make a match until Duncan’s prints were taken a second time months
later.
                         DUNCAN v. ORNOSKI                         7315
was appointed as defense counsel for Duncan’s trial. Based on
Cheroske’s fee requests, Duncan’s habeas counsel calculated
that Cheroske spent no more than 35.1 hours preparing for
Duncan’s capital murder trial. People v. Duncan, 

, 135 (Cal. 1991).4 During the trial, evidence was intro-
duced to show that Duncan was present at the crime scene,
including his palm prints, fingerprint, and shoe print that were
found at the scene. The jury also heard testimony from Greg-
ory Matheson, a criminologist for the Serology Section of the
Los Angeles Police Department, regarding blood found at that
location. Matheson explained how he examined the blood
lifted from the scene and compared his results with DeBaun’s
bloodtype, which was Type O, and genetic markers.5 Three
samples that he tested were inconsistent with DeBaun’s blood
—Items No. 5, 8, and 10.

   Item No. 5 is a cloth square that was used to lift a blood
stain from a partial shoe print left in blood on the money room
floor. Matheson testified that Item No. 5 tested positive for A,
B, and H antigens. When asked by the prosecutor whether it
is possible for something other than blood to display that anti-
genic activity, Matheson explained that bacteria, animal
  4
     This figure does not include time Cheroske spent during the prelimi-
nary hearing as retained counsel.
   5
     The principal system used to categorize human blood is the A.B.O.
system, which is made up of four blood types: Type O, Type A, Type B,
and Type AB. Blood type is determined by testing for the presence of mol-
ecules called “antigens.” All blood contains H antigens. People with Type
A blood also possess A antigens. Those with Type B blood also possess
B antigens. Those with Type AB blood also possess both A and B anti-
gens. A person with Type O blood possesses only H antigens. A person’s
A.B.O. blood type is genetically determined and never changes throughout
his life. The victim had Type O blood. Duncan also has Type O blood.
Therefore, any antigens other than H that came from human bodily fluid
necessarily came from someone other than Duncan or the victim. There
are additional genetic markers that can be used to categorize blood. The
presence or absence of certain enzymes determines a person’s various
blood subtypes.
7316                      DUNCAN v. ORNOSKI
blood, or other contaminants could have produced those
results.

   Item No. 8 is a blue and white Handiwipe rag with red
stains that was removed from a table in the money room. The
table was located below a first aid kit that hung on the wall.
The first aid kit was found open and first aid supplies were
missing, which led detectives to speculate that the killer was
injured during the attack and used some of the first aid to treat
himself. The rag was found lying on top of the padlock that
was used to secure the key box where the key to the looted
supervisor’s bank was kept. Matheson testified that “there
was A.B.O. activity that was present [on the rag] that was dif-
ferent from Miss DeBaun’s but the bloodstain still could have
come from her, but have been contaminated by some other
source.”

   Item No. 10 is a cloth square used to remove a red stain
from the floor.6 Matheson testified that he detected A and B
antigens, but no activity that was consistent with the victim’s
blood. Matheson concluded that this could mean that Type A
and Type B blood or Type AB blood might be present.

  Later in the trial, the State recalled Matheson to the stand
and he testified that the stain on Item No. 10 was “mostly
gray” instead of red or brown. This was inconsistent with the
way he would “normally find a good bloodstain to be.” In his
supplemental testimony, Matheson also stated that Item No.
10 tested as a “weak positive” in the presumptive blood test
and a “weak human positive” in the human species test. These
  6
    There is some dispute as to where Item No. 10 was found. Detective
Lewellen’s Property Report describes the stain as located “on E side of
door leading from store room to bar.” The money room is east of the bar
so if the property report is correct, the stain would have been lifted from
just inside the money room. At trial, however, Detective Lewellen testified
that the stain was lifted from the floor of the bar just outside the money
room. There are no further clues in the record as to where the blood was
actually found.
                      DUNCAN v. ORNOSKI                    7317
results caused him to doubt whether the stain actually con-
tained human blood. Matheson further testified that Item No.
10 did not test positive for any of the different sub-types cal-
led genetic markers. According to Matheson, a number of
things can cause this, including the fact that the sample was
old, that the sample was not blood, or that it was a fresh sam-
ple that was too weak to produce a typing result. Matheson
declined to draw any conclusions about the antigenic activity
on Item No. 10. He stated: “All I know is that I got a weak
indication of blood, and I did get antigenic activity. It could
be from the blood.”

   In addition to the three samples described above, blood
samples were taken from underneath the victim’s fingernails.
Matheson testified at trial that these blood samples were Type
O blood and contained a subtype that was consistent with the
victim’s blood.

   Matheson could not testify about whether the blood sam-
ples on any of the three items or from the fingernail scrapings
could have belonged to Duncan because he was not given a
sample of Duncan’s blood. In fact, Duncan’s blood type was
unknown at the time of trial so it could not be determined
whether the A and B antigens that were found at the crime
scene belonged to him.

   When Cheroske cross-examined Matheson, his lack of
preparation was evident. At the outset, he said to Matheson:
“[Y]ou lost me . . . when you’ve been talking about all of
these numbers.” Moreover, his line of questioning highlighted
what turned out to be a damaging issue to the defense.
Cheroske asked Matheson: “Now, taking Number 10 first, if
you were given a whole blood sample from someone else,
let’s say this man over here, Mr. Henry Duncan, could you
analyze it and tell if they are different types?” Matheson
replied that he could have done so. Cheroske then asked
Matheson twice whether he had compared the blood found at
the scene with any other whole blood samples besides the vic-
7318                  DUNCAN v. ORNOSKI
tim’s. Matheson replied that he had not been given any other
blood samples besides the victim’s. Cheroske’s cross-
examination only served to emphasize the fact that although
Matheson could have determined whether Duncan’s blood
type was consistent—or inconsistent—with the blood found at
the crime scene that did not belong to the victim, Duncan’s
blood was never given to the serologist to test. The prosecutor
turned this fact to his advantage in his closing argument.
Cheroske did not present expert witness testimony from a
serologist or any other type of expert.

   Cheroske’s defense theory was that someone other than
Duncan committed the murder. Detective Leroy Orozco testi-
fied at trial that, “because of the smallness of the area of the
money room, that it’s possible for two suspects, but I believe
that only one could have attacked and killed Eileen.”
Cheroske argued that there was reasonable doubt as to
whether this single killer was Duncan, but he did not support
his theory with evidence that anyone other than Duncan was
present or that anyone other than Duncan was the actual
killer.

   Cheroske presented circumstantial evidence that a number
of suspicious individuals had been seen near the restaurant on
the night of the murder. He attempted to show that the perpe-
trator must have been an “outsider,” rather than an employee,
by presenting evidence that the victim was tortured before her
death, presumably in an attempt to get her to open a Brink’s
safe located inside the money room. He argued that the mur-
derer must not have known, as an employee such as Duncan
would have, that DeBaun could not open the safe without a
second key that only the Brink’s guard possessed. However,
as Cheroske elicited on cross-examination from a restaurant
employee, there was a sign on the front of the safe that read:
“This can only be opened by Brink’s.”

  During his closing argument, Cheroske argued that the
prosecutor should have had Duncan’s blood tested. The prose-
                       DUNCAN v. ORNOSKI                        7319
cutor responded in his closing argument that if the blood
found at the scene had been inconsistent with Duncan’s blood
type, Cheroske would have presented that evidence:

    Mr. Cheroske said we should have gotten blood from
    the defendant and we should have compared it. . . .
    But, don’t you think that the defense would have
    taken the blood of this defendant and brought it into
    court here with their own expert and had them testify
    to it if they thought there was the slightest chance to
    be able to prove that? . . . What would you do?
    Would you have the defendant’s blood taken by your
    own person so that the prosecution could never find
    out about it, have it tested, find out what it is, and if
    it’s different, you use it. If it’s the same you never
    use it.

The prosecutor’s argument implied that Cheroske actually did
have the blood tested and the reason Duncan’s blood type was
never presented to the jury was that it was consistent with the
blood found at the crime scene.

   On March 3, 1986, following the conclusion of trial, the
jury found Duncan guilty of first degree murder and robbery,
found the special circumstance to be true and found that Dun-
can murdered DeBaun with a knife. The next morning, a brief
penalty phase trial was held and the jury returned a death ver-
dict the following day.

  C.     State Appeals and Federal Habeas Corpus Review

    i.    California State Court

  On direct appeal, the California Supreme Court unani-
mously affirmed the verdict and the sentence. People v. Dun-
can, 

(Cal. 1991). However, in an unpublished
order the court subsequently denied Duncan’s petition and
supplemental petition for habeas corpus by a four to three
7320                  DUNCAN v. ORNOSKI
vote. The three dissenting justices would have issued an order
to show cause.

    ii.   Federal Habeas Corpus

   Duncan filed an original petition in the district court on
March 4, 1993 and filed the operative Second Amended Peti-
tion on December 10, 1996. The petition included twenty-nine
constitutional claims. Initially the district court ordered an
evidentiary hearing on Claims 5, 8, 12, 13 and 17. The case
was then stayed pending the resolution of Lindh v. Murphy,

(1997), in which the Supreme Court held that
the Antiterrorism and Effective Death Penalty Act of 1996
(“AEDPA”) does not apply to cases that were pending at the
time of its passage. 

   Instead of proceeding with the evidentiary hearing as previ-
ously proposed, the parties agreed that Claims 5, 8, and 13
could be decided on the state court record. In an order filed
on May 26, 1998, the district court denied Claim 8 (regarding
trial counsel’s misunderstanding of basic legal principles) and
Claim 13 (regarding ineffective assistance of trial counsel for
failure to object to certain statements made by the prosecutor
during the closing argument of the penalty phase) based
solely on the state court record.

   The court subsequently denied Claim 5, based solely on the
state court record, in an order filed on December 29, 1999. In
Claim 5, Duncan alleges that Cheroske rendered ineffective
assistance when he failed to consult a serologist and investi-
gate the potentially exculpatory blood evidence found at the
crime scene. Moreover, in that Claim, he asserts that the out-
come of his case was prejudiced by Cheroske’s failure to
present serological evidence to the jury that suggested that an
accomplice was present on the night of the murder.

   In the state court post-conviction proceeding, counsel had
arranged to have Duncan’s blood tested and the tests had
                      DUNCAN v. ORNOSKI                    7321
revealed that Duncan has Type O blood, the same type as the
victim’s. This means that the antigenic activity that Matheson
observed on Item Nos. 5, 8, and 10 that was inconsistent with
the victim’s blood could not have come from Duncan either.

   Also in the state court post-conviction proceeding, Mathe-
son, the State’s serologist, submitted a declaration about the
blood evidence and his trial testimony. He declared that if he
had been questioned further at trial, he would have stated his
conclusion that “the stain on Item No. 5 does represent human
blood.” Furthermore, Matheson stated in the declaration that
in his professional estimation, it is “more likely than not that
Item No. 5 represents one of the following: 1) a mixture of
Type AB blood and Type O blood; 2) a mixture of Type A,
Type B and Type O blood; 3) Type AB blood only; 4) a mix-
ture of Type A and Type B blood; 5) any other combination
of types providing the observed antigens; or 6) any combina-
tion of a blood sample mixed with other body fluids that exist
in any of the combinations described in examples 1-5 above.”
The A and B antigens could not have come from Duncan or
the victim because both individuals have Type O blood.
Matheson also testified that although “anything is possible,”
contamination of blood samples with chemicals that create
false positives “is not seen very often” and when it is, it usu-
ally happens on “clothing items, possibly due to the dyes or
other chemicals present.”

   About the sample from Item No. 8, Matheson declared that
“it is more likely than not that the stain on Item No. 8 is a
mixture of human body fluids that include Type O blood and
also body fluid(s) (such as blood, saliva or perspiration) from
either a Type AB individual or Type A and Type B individu-
als.” He further stated that it was “less likely” that a “non-
human source” contaminated the sample. This is because he
tested an unstained portion of the rag to verify that the anti-
gens he detected were not coming from the rag itself. That
control did not show any A.B.O. activity. According to
Matheson,“[t]his represents a strong indication that the A and
7322                  DUNCAN v. ORNOSKI
B antigens on Item No. 8 are in fact from a human source.”
Again, the A and B antigens could not have come from either
Duncan or the victim.

   Finally, with regard to Item No. 10, Matheson declared that
“though there is some doubt about the stain because of its
color, in my expert opinion it is more likely than not that the
stain on Item No. 10 does represent body fluid(s) from an
individual or individuals with A and B antigens.” Moreover,
Edward Blake, a serology expert hired by Petitioner,
explained in his declaration that A and B antigens often test
strongly even when they are present in a small sample. If a
small amount of blood was present on Item No. 10, it would
make sense for the A and B antigens to test strongly even
though there was only a weak indication of human blood and
other genetic markers. Thus, the results of Matheson’s testing
of Item No. 10 are not unusual and are consistent with the
conclusion that Item No. 10 contained human blood.

   Moreover, Duncan’s blood test revealed that although he
has the same blood type as the victim, he has different genetic
subtypes. Matheson testified at trial that the blood taken from
underneath the victim’s fingernails was of the same genetic
subtypes as the victim’s blood. Thus, the blood in those sam-
ples could not have belonged to Duncan. According to Mathe-
son’s Analyzed Evidence Report, many of the blood samples
that he tested did not yield conclusive results for genetic sub-
types. For all of the samples that did yield such results, the
genetic subtype results matched those of the victim and not
those of Duncan. Although the Type O blood from the crime
scene with undetermined genetic subtypes could have
belonged to Duncan, none of the Type O blood for which the
genetic subtypes could be determined did belong to him.

   Both Blake, Duncan’s habeas serology expert, and Mathe-
son summarized their conclusions about the blood evidence at
the crime scene. According to Matheson:
                      DUNCAN v. ORNOSKI                       7323
    All things considered, the most likely explanation of
    the evidence is that there was human body fluid at
    the crime scene that came from at least one (person
    who did not have Type O blood, and that that body
    fluid was deposited at the crime scene after the last
    time the floor was washed.

  Blake’s conclusion was as follows:

    Based on the evidence, on the tests and controls per-
    formed by Mr. Matheson, and on my thirteen years
    as an expert serologist, I believe that there is only
    one reasonable and non-speculative interpretation of
    Mr. Matheson’s results and of the other evidence: In
    my professional opinion, on the night of November
    13, 1984, at some time after the restaurant was
    closed to the public, and after the routine floor wash-
    ing was completed, someone with Type AB blood
    was injured, shed blood on the money room floor,
    and wiped off some of his or her blood on the blue
    and white Handiwipe rag. That person was neither
    the victim Josephine DeBaun nor the petitioner
    Henry Earl Duncan.

   Both experts concluded that the blood samples found on the
floor of the crime scene were, in all likelihood, deposited after
the last time the floor was washed. Because the significance
of the floor mopping was not known until after Duncan’s
blood was tested, neither party presented evidence on the
question at trial. During the state court post-conviction pro-
ceedings, Peter Feimann, a bartender at the Host International
Restaurant, declared that it was a regular and customary prac-
tice to mop the floors every night with soap and water. A
manager, Shirley Goodell, stated that the floor mopping was
not always thorough. Blake clarified, however, that even a
“routine washing of the floor” would be “strongly likely to
remove measurable antigens from the floor.”
7324                  DUNCAN v. ORNOSKI
   The State offered evidence that the night shift was short
one utility worker on the night of the murder. However, Fei-
mann stated that the person responsible for mopping the
floors was working that night. Additionally, soon after the
murder, the police interviewed two witnesses, Mark A. Chris-
tian and Arnell E. Jackson, who, separately and without
prompting, claimed that they saw a custodian mopping the
floor of the restaurant on the night of the murder. Cheroske
had copies of the police reports in which Christian and Jack-
son were quoted, but he did not call either of them to testify
at trial.

   It is undisputed that Cheroske never had Duncan’s blood
tested to determine its genetic markers. In a sworn declara-
tion, Cheroske stated that the reasons he did not test Duncan’s
blood were (1) Duncan admitted to him that he had been pres-
ent in the money room and (2) he did not want to risk further
tying Duncan to the scene.

   After considering the record of the state proceeding that
contained the aforementioned declarations and expert testi-
mony regarding the floor washing and the blood found on
Item Nos. 5, 8, and 10, the district court held that the evidence
presented was not enough to establish ineffective assistance of
counsel. The court found that when Cheroske decided not to
test Duncan’s blood, he reasonably relied on Duncan’s state-
ment that he was present during the murder. The court also
found that Duncan was not prejudiced by Cheroske’s omis-
sions.

  Next, the district court granted the State’s motion for sum-
mary judgment on the remaining guilt phase and most of the
penalty phase claims. It ordered an evidentiary hearing, how-
ever, on the portions of Claims 9 and 17 regarding ineffective
assistance of counsel “based on trial counsel’s failure to
request funding for, or to consult with, a drug expert, or offer
expert testimony on drug addiction at the penalty phase of
petitioner’s trial.”
                      DUNCAN v. ORNOSKI                    7325
   At the evidentiary hearing, the parties submitted deposition
testimony, declarations, and other evidence, and the court
heard some live testimony. Duncan presented evidence from
several experts regarding the professional norms in 1985 and
1986 of investigating and presenting mitigating evidence at
the penalty phase of a capital case. He also presented addi-
tional testimony from lay witnesses about his background,
exposure to traumatic events in his childhood, and positive
characteristics. A psychologist prepared and submitted a
social history of Duncan. He and other experts testified about
the psychological effects on Duncan of his childhood traumas
and drug use, his personality disorders, his genetic predisposi-
tion to substance abuse, and indications that he suffered
organic brain damage. The State presented the testimony of its
own experts who challenged some of the conclusions of Dun-
can’s experts.

  The district court held that Cheroske’s failure to consult an
expert on drug use was deficient, but that his deficient perfor-
mance did not prejudice Duncan. It then denied Duncan’s
Second Amended Petition for Writ of Habeas Corpus in its
entirety in a judgment filed on September 30, 2005. Duncan
now appeals.

II.   Standard of Review

   [1] The claims that Duncan asserts on appeal were made in
his federal habeas petition that was filed prior to the enact-
ment of AEDPA, so the provisions of that Act do not apply
to his claims. See Lindh v. 

;
Douglas v. Woodford, 

, 1085 (9th Cir. 2003).
This court reviews the district court’s decision to deny habeas
relief de novo. Lambright v. Schriro, 

, 1113-14
(9th Cir. 2007). Ineffective assistance of counsel claims are a
mixed question of law and fact and we review them de novo.
Summerlin v. Schriro, 

, 628 (9th Cir. 2005) (en
banc). We review for clear error the district court’s findings
of fact. Frierson v. Woodford, 

, 988 (9th Cir.
7326                  DUNCAN v. ORNOSKI
2006). Finally, because this case is not subject to the provi-
sions of AEDPA, “we do not review the state court’s legal
conclusions to determine whether they are ‘objectively unrea-
sonable;’ rather, we ‘simply resolve the legal issue on the
merits, under the ordinary rules.’ ” 

(quoting Belmontes v. Brown, 

, 1101
(9th Cir. 2005), rev’d on other grounds, Ayers v. Belmontes,

(2006)).

III.   Discussion

   In order to convict Duncan of first degree felony murder,
the State had to prove only that he intended to commit the
robbery and that the murder occurred in the course of the rob-
bery. Because there was compelling evidence, including the
palm prints, fingerprint, and shoe print, that Duncan was pres-
ent and participated in the robbery, and because the murder
was clearly related to the robbery, we reject Duncan’s claim
that his conviction should be reversed on the ground of inef-
fective assistance of counsel. Counsel’s failure to obtain a
serology expert and to have blood tests performed was harm-
less with respect to the conviction. In order to prove the spe-
cial circumstance allegation that would trigger the penalty
phase, however, the State was required to show that Duncan
himself intentionally killed DeBaun or that he intended that
she be killed by an accomplice. Here, the serological evidence
that Cheroske failed to develop during trial would have raised
substantial doubt as to whether Duncan actually killed
DeBaun. Moreover, if he was not the killer, the evidence in
the record would not have established beyond a reasonable
doubt that he intended that DeBaun be killed. Thus, the spe-
cial circumstance allegation is the focus of our analysis.

   There was significant evidence that an individual other than
the victim was wounded and shed blood in the money room
on the night of the crime. Because there was evidence of a
struggle during the course of the attack, and the injury proba-
bly resulted from this struggle, the killer was likely the donor
                       DUNCAN v. ORNOSKI                   7327
of the blood that did not belong to the victim. Moreover, the
State advanced the theory, supported by the testimony of a
police detective, that the money room was too small to
accommodate more than one killer. Thus, if Cheroske could
show that some of the blood in the money room was neither
DeBaun’s nor Duncan’s, considerable doubt would be raised
that Duncan was the killer; if he were not, there is little, if
any, evidence that would tend to show that he intended that
DeBaun be killed. Accordingly, Cheroske’s failure to investi-
gate and present the serological evidence prejudiced Duncan
with respect to the jury’s special circumstance finding.

  A.     Ineffective Assistance of Counsel

   [2] A defendant’s Sixth Amendment right to representation
in a criminal trial includes “the right to the effective assis-
tance of counsel.” McMann v. Richardson, 

, 771
n.14 (1970). To prevail on a claim of ineffective assistance,
a petitioner must show that: (1) his trial counsel’s perfor-
mance “fell below an objective standard of reasonableness”;
and (2) “there is a reasonable probability that, but for coun-
sel’s unprofessional errors, the result of the proceeding would
have been different.” Strickland v. Washington, 

,
688, 694 (1984).

   Because Cheroske’s failure to consult an expert in serology
that would have enabled him to make informed, strategic
decisions with regard to the blood evidence, his failure to
have Duncan’s blood tested, and his failure to present the
results of those tests at trial were constitutionally deficient,
and because his deficient performance casts substantial doubt
on the jury’s special circumstance finding, we reverse the dis-
trict court’s denial of the writ of habeas corpus.

    i.    Deficient Performance

  Under Strickland, we must presume that counsel was com-
petent and Duncan must rebut this presumption by showing
7328                  DUNCAN v. ORNOSKI
that his performance was objectively unreasonable under pre-
vailing professional norms and was not the product of sound
strategy. 

scrutiny of counsel’s perfor-
mance must be highly deferential,” and we must evaluate
counsel’s conduct from his perspective at the time, without
the benefit of hindsight. 

“[S]trategic choices made
after thorough investigation of law and facts relevant to plau-
sible options are virtually unchallengeable.” 

How-
ever, decisions that are made before a complete investigation
is conducted are reasonable only if the level of investigation
was also reasonable. “[C]ounsel has a duty to make reason-
able investigations or to make a reasonable decision that
makes particular investigations unnecessary. In any ineffec-
tiveness case, a particular decision not to investigate must be
directly assessed for reasonableness [under] all the circum-
stances . . . .” 

see also Wiggins v. Smith, 

, 521-22 (2003).

   [3] The Supreme Court has “declined to articulate specific
guidelines for appropriate attorney conduct and instead [has]
emphasized that ‘[t]he proper measure of attorney perfor-
mance remains simply reasonableness under prevailing pro-
fessional norms.’ ” 

(quoting

). However, we have established
general principles that guide our determination of what consti-
tutes objectively reasonable attorney performance, including
the duty to investigate. See 

.

   [4] This court has repeatedly held that “[a] lawyer who fails
adequately to investigate and introduce . . . [evidence] that
demonstrate[s] his client’s factual innocence, or that raise[s]
sufficient doubt as to that question to undermine confidence
in the verdict, renders deficient performance.” Hart v. Gomez,

, 1070 (9th Cir. 1999) (holding that counsel’s
failure to review key documents corroborating defense wit-
ness’s testimony constituted deficient performance); see also
Avila v. Galaza, 

, 919 (9th Cir. 2002) (holding
that counsel’s failure to investigate evidence that defendant’s
                      DUNCAN v. ORNOSKI                    7329
brother was the shooter constituted deficient performance);
Lord v. Wood, 

, 1095-96 (9th Cir. 1999) (hold-
ing that counsel’s failure to call key witnesses whose testi-
mony undermined the prosecutor’s case constituted deficient
performance); Sanders v. Ratelle, 

, 1457 (9th
Cir. 1994) (holding that counsel’s failure to investigate evi-
dence that someone else was the killer constituted deficient
performance). “The failure to investigate is especially egre-
gious when a defense attorney fails to consider potentially
exculpatory evidence.” Rios v. Rocha, 

, 805 (9th
Cir. 2002); see also Harris v. Wood, 

, 1435-37
(9th Cir. 1995) (holding that counsel’s failure to retain an
investigator and interview many of the individuals identified
in the police reports was deficient performance).

   We allow lawyers considerable discretion to make strategic
decisions about what to investigate, but only after those law-
yers “have gathered sufficient evidence upon which to base
their tactical choices.” Jennings v. Woodford, 

,
1014 (9th Cir. 2002). When defense counsel merely believes
certain testimony might not be helpful, no reasonable basis
exists for deciding not to investigate. See 

(holding that the lawyer’s belief that certain testimony
might not be helpful at trial was an unreasonable basis upon
which to decide not to investigate and thus constituted defi-
cient performance).

   [5] Although it may not be necessary in every instance to
consult with or present the testimony of an expert, when the
prosecutor’s expert witness testifies about pivotal evidence or
directly contradicts the defense theory, defense counsel’s fail-
ure to present expert testimony on that matter may constitute
deficient performance. See Caro v. Woodford, 

,
1254-56 (9th Cir. 2002) (holding that counsel was deficient
for failing to consult an expert and present expert testimony
about the physiological effect of toxic chemical exposure on
defendant’s brain); Miller v. Anderson, 

, 459
(7th Cir. 2001) (finding deficient performance when counsel
7330                  DUNCAN v. ORNOSKI
failed to hire an expert to rebut the prosecution’s expert testi-
mony about physical evidence linking defendant to the crime
scene when the defense theory was that defendant was not at
the crime scene), remand order modified by stipulation, 

(7th Cir. 2001) (vacated at request of parties when
settlement was reached); Troedel v. Wainwright, 

, 1461 (S.D. Fla. 1986) (holding that counsel’s failure to
depose the State’s expert, and more important, failure to con-
sult with an expert in order to contradict key evidence of the
“most crucial aspect of the trial” was deficient), aff’d Troedel
v. Dugger, 

(11th Cir. 1987) (per curiam).

   [6] Here, Cheroske’s defense was that Duncan did not kill
DeBaun, yet he did not advance any plausible alternative the-
ory or present any specific evidence that he was not the mur-
derer. There was, however, specific evidence that was before
the jury that could have shown that Duncan did not kill
DeBaun—the blood samples. Despite the fact that Cheroske
had a copy of the police serology report and that his highlight-
ing and underlining of that report suggest that he understood
the significance of the blood samples that contained antigenic
activity that was inconsistent with the victim’s blood,
Cheroske did not consult a serology expert or have Duncan’s
blood tested. Cheroske’s failure to consult a serologist when
there existed potentially exonerating blood evidence, and his
subsequent failure to have Duncan’s blood tested and present
the results of those tests at trial were unreasonable under pre-
vailing professional norms.

   [7] Cheroske did not consult a serologist or conduct any
investigation whatsoever with regard to the blood evidence. It
is especially important for counsel to seek the advice of an
expert when he has no knowledge or expertise about the field.
See Dugas v. Coplan, 

, 331 (1st Cir. 2005).
Cheroske demonstrated his lack of expertise in serology at the
outset of his cross-examination of the State’s serology expert
when he told him, “You lost me.” Moreover, it appears that
Cheroske did not even know what serology was. Serology is
                      DUNCAN v. ORNOSKI                     7331
the study of blood serum, yet when Cheroske cross-examined
Matheson, a serology expert, he asked him about hair evi-
dence. Clearly, Cheroske did not have the personal expertise
in serology to make strategic decisions about how to handle
the blood evidence on his own and he certainly was not quali-
fied to undermine the State’s case by simply cross-examining
its experts without obtaining expert assistance himself.

   [8] Additionally, the central role that the potentially excul-
patory blood evidence could have played in Duncan’s defense
increased Cheroske’s duty to seek the assistance of an expert.
The samples of blood that did not belong to the victim were
the only forensic evidence that had not been linked to Duncan
and that could have established that someone other than he
was the murderer. Whereas the State’s experts testified that
the fingerprint, palm prints, and shoe prints that were found
at the scene likely belonged to Duncan, there was no expert
who could link Duncan to the blood. Cheroske should have
consulted a serology expert as soon as he read the evidence
report and noticed that there was blood found at the crime
scene that did not belong to the victim so that he could fully
understand the forensic implications of that evidence and
make an informed decision about how to proceed at trial. See
Driscoll v. Delo, 

, 709 (8th Cir. 1995) (“[A] rea-
sonable defense lawyer would take some measures to under-
stand the laboratory tests performed and the inferences that
one could logically draw from the results.”).Cheroske pro-
vided no explanation for why he failed to consult a serology
expert or investigate the potentially exculpatory blood evi-
dence. He certainly did not advance a strategic or tactical rea-
son for failing to do so. The record, however, reveals one
possible explanation that appears in his argument in opposi-
tion to a motion that the State filed during trial to compel a
blood test to determine Duncan’s blood type. Cheroske stated
that in reading the evidence reports, it was “clear to me that
the typings were different and it was not my burden to do any-
thing with regard to that other than in all fairness to rely on
the state of the evidence as it was produced.” Apparently,
7332                      DUNCAN v. ORNOSKI
Cheroske did not believe that when he became aware of the
blood evidence, he had any duty to consult an expert who
could assist him in preparing his cross-examination and serve
as an expert witness on Duncan’s behalf, or make any effort
to establish that the blood that came from someone other than
the victim did not belong to Duncan. Cheroske’s inaction was
unreasonable, especially given that the blood samples were
the only potentially exculpatory evidence in the case.

   [9] Had Cheroske consulted a serologist, he would have
been prepared to make the necessary decisions regarding the
testing of Duncan’s blood, what further investigation should
be done, and how to approach his cross-examination of the
State’s serology expert. Without consulting an expert,
Cheroske had no basis upon which to devise his defense strat-
egy. His failure to investigate the blood evidence, despite the
fact that he knew that it existed and that he understood the
possible implications of it, constituted deficient performance.

   Cheroske’s failure to have Duncan’s blood tested in order
to determine whether the blood that was found at the crime
scene that did not belong to the victim was his also consti-
tuted deficient performance. Having Duncan’s blood tested
posed no risk to Duncan’s defense, but the potential benefit
was enormous. The blood evidence, detailed in the police
report that Cheroske highlighted and underlined, was the only
forensic evidence that had not yet been tied to Duncan. Thus,
it presented Cheroske with his only opportunity to cast doubt
on the State’s theory that Duncan was the actual killer.

   The district court held that Cheroske’s failure to have Dun-
can’s blood tested was not deficient because it was based on
a reasonable tactical decision that is entitled to deference. The
court found that Cheroske decided not to have Duncan’s
blood tested because Duncan admitted to him that he had been
present in the murder room and Cheroske did not want to fur-
ther connect Duncan to the crime scene.7 The district court
clearly erred in its holding.
  7
   The district court went on to speculate that counsel could have reason-
ably decided that investigating the blood evidence would be fruitless given
                           DUNCAN v. ORNOSKI                           7333
   [10] Cheroske’s first explanation for failing to have Dun-
can’s blood tested was that Duncan told him that he was
“present in the murder room.” However Duncan had also told
his counsel that “he was neither the killer nor did he aid in the
killing.” The only scenario in which those two statements
could be true is one involving an accomplice who was the
actual killer. Blood at the crime scene that belonged to neither
Duncan nor to the victim would have been evidence not only
that there was someone else involved in the attack, but also
that the other perpetrator was likely the actual killer, not the
bystander who was merely “present,” given that the actual
killer would have been the one injured in a struggle. When
“tantalizing indications in the record” exist, such as the poten-
tially exculpatory blood evidence in this case, a reasonable

the rest of the evidence tying Duncan to the crime. The State further con-
tends that Cheroske told a prosecutor that he did not test Duncan’s blood
because Duncan refused to take a blood test and Duncan revealed to
Cheroske that he had received cuts during the crime so that any blood at
the crime scene would have been his. The district court noted the State’s
argument and acknowledged that those reasons would also support coun-
sel’s decision not to investigate the blood evidence, but the court did not
find that these were Cheroske’s actual reasons, nor did it find that
Cheroske actually made the statements that the State alleged. In light of
the Supreme Court’s admonitions that reviewing courts may not substitute
their own strategic reasoning for that of trial counsel in order to find that
counsel’s performance was justified, we do not consider these additional
speculative justifications to be Cheroske’s actual reasons for declining to
test Duncan’s blood. See 

(“[T]he ‘strategic
decision’ the state courts and respondents all invoke to justify counsel’s
limited pursuit of mitigating evidence resembles more a post hoc rational-
ization of counsel’s conduct than an accurate description of their delibera-
tions prior to sentencing.”); see also Alcala v. Woodford, 

,
871 (9th Cir. 2003) (“We will not assume facts not in the record in order
to manufacture a reasonable strategic decision for Alcala’s trial counsel.”);
United States v. Burrows, 

, 918 (9th Cir. 1989) (per curiam)
(holding trial counsel deficient for failing to investigate an insanity
defense partly because “the district court’s assumptions that the attorney
must have considered an insanity defense and might have rejected it for
strategic reasons appear not to have been based on the record”).
7334                  DUNCAN v. ORNOSKI
attorney would investigate further. Stankewitz v. Woodford,

, 720 (9th Cir. 2004).

   [11] Cheroske’s second stated reason for not having Dun-
can’s blood tested was that given the fingerprint evidence, he
did not want to risk further tying Duncan to the crime scene.
This explanation is unpersuasive. It reveals Cheroske’s igno-
rance about forensic evidence and further demonstrates why
he should have consulted an expert before making critical tac-
tical decisions. As the State’s fingerprint expert, George Her-
rera, testified, “nobody in the world has the same
fingerprints,” therefore, fingerprints and palm prints can be
ascribed to a specific individual with certainty. Given that one
of Duncan’s fingerprints and three of his palm prints were
found at the crime scene, he was clearly present. If Cheroske
had consulted a serology expert, he would have learned that
blood evidence cannot be tied to a specific individual. Blood
evidence can only be used to rule out a certain percentage of
the population as potential donors of a particular sample.
Thus, there was little risk of further tying Duncan to the crime
through the blood evidence. Cheroske did not have the requi-
site knowledge or information to make any decisions about
the blood evidence without consulting an expert, so his failure
to have Duncan’s blood tested is not entitled to deference. See

(“[C]ounsel can hardly be said to have
made a strategic choice when s/he [sic] has not yet obtained
the facts on which a decision could be made.”) (quoting Sand-

(internal quotation marks omitted) (quot-
ing United States v. Gray, 

, 711 (3d Cir. 1989)).

   Moreover, there would have been no harm in getting Dun-
can’s blood tested, even if the tests had revealed that his blood
was consistent with the blood in items 5, 8, and 10, which
they did not, because Cheroske could have determined Dun-
can’s blood type without having to reveal the result. At the
time of trial, California law permitted confidential testing by
defense experts. Cal. Penal Code § 987.9 (West 1986). Even
the prosecutor advised the jury in his closing argument that
                       DUNCAN v. ORNOSKI                     7335
Cheroske could have had Duncan’s blood tested in secret. In
fact, the prosecutor took full advantage of the fact that Dun-
can’s blood type was never presented at trial and suggested to
the jury that the omission proved that the blood evidence was
inculpatory. He told the jury that Cheroske likely would have
tested Duncan’s blood in secret but would have revealed the
results only if they were favorable to him. He then argued to
the jury that the reason that there was no evidence presented
at trial about Duncan’s blood type was that the evidence was
consistent with the theory that Duncan was the killer.

   Moreover, even if Cheroske feared that California’s proce-
dures for getting a sealed court order to test Duncan’s blood
would not have protected the confidentiality of the results, he
could have determined Duncan’s blood type surreptitiously
without a court order. If Cheroske had consulted a serology
expert, he would have known that most of the population’s
blood type can be detected in bodily fluid other than blood.
Duncan is one such person, a so-called “secretor,” so
Cheroske could have obtained a small sample of his saliva in
a vial or cloth and used that to determine his blood type with-
out notifying the court or the State. Cheroske had nothing to
lose by testing Duncan’s blood, but he stood to gain crucial
evidence by doing so.

   The State contends that Cheroske did not test Duncan’s
blood because Duncan admitted to him that he had committed
the murder. The district court did not find this to be
Cheroske’s reason, nor does the record support such a claim.
Even if this were Cheroske’s actual reason, however, it would
be unreasonable. A defendant’s admission of guilt to his law-
yer does not absolve the lawyer of his duty to investigate the
crime. The professional norms in existence at the time of
Duncan’s trial and recognized by the Supreme Court clearly
state that counsel must “explore all avenues leading to facts
relevant to the merits of the case . . . . The duty to investigate
exists regardless of the accused’s admissions or statements to
the lawyer of facts constituting guilt . . . .” Rompilla v. Beard,
7336                  DUNCAN v. ORNOSKI

, 387 (2005) (quoting 1 ABA Standards for
Criminal Justice 4-4.1 (2d ed. 1982 Supp.) (internal quotation
marks omitted)).

   Strategic decisions based on information provided by the
defendant are often reasonable and entitled to deference. See,
e.g., 

(“Counsel’s actions are usu-
ally based, quite properly, . . . on information supplied by the
defendant.”). However, counsel must consider all of the
defendant’s statements, not just those that make his job easier.
Even if we were to give credence to the State’s dubious alle-
gation that Duncan confessed to Cheroske, which we do not,
the glaring inconsistencies in Duncan’s reported accounts of
the murder made it unreasonable for Cheroske to rely on any
one of Duncan’s statements in isolation when making tactical
decisions about investigating the crime. Moreover, the fact
that having Duncan’s blood tested could not possibly have
harmed the case, even if the results were inculpatory, because
the test results could have remained confidential makes
Cheroske’s failure to do so even more unreasonable.

   The State also contends that Duncan refused to have his
blood tested. The district court did not find this to be
Cheroske’s reason for not having the test performed and
Cheroske does not cite this as his reason in his sworn declara-
tion. Even assuming that it were true, however, Cheroske
could have attempted to discover Duncan’s blood type in
other ways. For example, he could have examined Duncan’s
medical records to see if they contained his blood type.
Cheroske never obtained those records. As we have previ-
ously held, “if a client forecloses certain avenues of investiga-
tion, it arguably becomes even more incumbent upon trial
counsel to seek out and find alternative sources of information
and evidence, especially in the context of a capital murder
trial.” Silva v. Woodford, 

, 847 (9th Cir. 2002).
Moreover, had Cheroske consulted an expert, he could have
explained to Duncan that his blood type probably could be
                      DUNCAN v. ORNOSKI                    7337
determined from his saliva without the prosecutor discovering
the results, and Duncan might have cooperated.

   [12] None of the reasons given by Cheroske in his own
declarations or attributed to Cheroske by the State are consis-
tent with a sound defense strategy entitled to the deference of
this court. All of his stated reasons amount to a fear that the
blood samples found at the scene would be consistent with
Duncan’s blood. This fear was unwarranted for two reasons.
First, blood cannot be directly tied to its donor through blood
typing, which Cheroske would have known had he consulted
a serology expert. Thus, even if the blood samples were con-
sistent with Duncan’s blood type, there would have been no
way to prove that they came from him, unlike the fingerprint
and palm prints that were lifted from the crime scene and con-
clusively proved that Duncan was present. Second, Cheroske
could have tested Duncan’s blood without the knowledge of
the prosecutor and, once he knew the results, he could have
made a strategic decision about how to deal with the blood
evidence at trial. Cheroske’s failure to consult a serology
expert and determine Duncan’s blood type when he was on
notice of the existence of potentially exculpatory blood evi-
dence was unreasonable under prevailing professional norms
and constitutes deficient performance under Strickland.

    ii.   Prejudice

   [13] To establish that Cheroske’s deficient performance
prejudiced him, Duncan must show that there is “a reasonable
probability that, but for counsel’s unprofessional errors, the
result of the proceeding would have been different. A reason-
able probability is a probability sufficient to undermine confi-
dence in the outcome.” 

. It is not
necessary, however, to show that counsel’s deficient conduct
“more likely than not altered the outcome in the case.” Sand-

. In determining whether a defendant was
prejudiced by counsel’s inadequate representation, we exam-
ine the evidence that could have been presented to the jury
7338                  DUNCAN v. ORNOSKI
had counsel performed competently and compare that to the
evidence that the jury actually heard. See Bonin v. Calderon,

, 834 (9th Cir. 1995). If the difference between
the evidence that could have been presented and that which
actually was presented is sufficient to “undermine confidence
in the outcome” of the proceeding, the prejudice prong is sat-
isfied. 

.

   Before considering the question of prejudice, we will reiter-
ate the California procedure in death penalty cases, so as to
make it easier to understand to which portion of the proceed-
ings Cheroske’s deficient performance was prejudicial and to
which it was not. Under California law, there are three basic
decisions that a jury must make before the court may sentence
to death a defendant who is charged with first degree murder
with special circumstance. After the guilt phase of the trial is
complete, the jury must decide whether to convict the defen-
dant of first degree murder. Cal. Penal Code § 189 (West
1986). If it finds the defendant guilty of first degree murder,
it must next decide whether one or more of the alleged “spe-
cial circumstances” is true beyond a reasonable doubt. Cal.
Penal Code §§ 190.2, 190.4 (West 1986). If the jury finds the
defendant guilty of first degree murder and finds that one or
more of the special circumstances are true, then a penalty
phase trial is conducted, at the conclusion of which the jury
makes its final decision—whether to recommend that the
defendant be sentenced to life in prison without the possibility
of parole or to death.

   As we have stated earlier, even with the benefit of the addi-
tional blood evidence that Cheroske failed to present, we
would be required to affirm Duncan’s conviction for first
degree murder. There was sufficient evidence, in the form of
a fingerprint, shoe prints, and palm prints, to place him at the
scene of the crime on the night of the murder and to prove
that he intended to rob the restaurant. Because the murder was
committed in furtherance of the robbery, Duncan would be
guilty of first degree murder under California’s felony murder
                      DUNCAN v. ORNOSKI                     7339
rule even if he did not actually murder DeBaun or intend that
she be killed. See People v. Pulido, 

, 1239 (Cal.
1997).

   [14] The blood evidence becomes relevant, however, when
we consider the jury’s special circumstance finding. The State
alleged the special circumstance of robbery-murder, which, at
the time of trial, required a finding that the defendant intended
that the victim be killed, whether he was the actual killer or
an accomplice. See Carlos v. Superior Court, 

,
153-54 (1983), partially overruled by People v. Anderson, 

(1987) (holding that proving intent to kill is not
required when the defendant is the actual killer, but it is
required when the defendant is an accomplice). Consequently,
unlike the first degree murder conviction for which the jury
was required to find only that Duncan intended to rob the res-
taurant and that DeBaun was somehow killed in the course of
the robbery, the special circumstance rule required the jury to
find that Duncan intentionally killed DeBaun himself or, if
she was killed by an accomplice, that Duncan intended that
she be killed.

   Once Duncan’s presence at the crime scene was established
through the forensic evidence, the one-killer theory advanced
at trial and uncontradicted by defense counsel, coupled with
the brutal nature of the attack, compelled the finding that he
intentionally killed DeBaun. However, if Cheroske had pre-
sented the alternative theory that Duncan had an accomplice,
he could have raised a question in the minds of the jury as to
whether Duncan was the actual killer and whether he intended
that DeBaun be killed. As the district court acknowledged,
“[i]ntent to kill would have been harder to prove if petitioner
was not the actual killer.” If at least one juror had a reason-
able doubt that Duncan killed DeBaun or that he intended his
accomplice to kill her, the jury could not have returned a ver-
dict against Duncan on the special circumstance allegation.

  [15] Had Cheroske consulted a serologist, investigated the
blood evidence, and properly cross-examined the State’s wit-
7340                  DUNCAN v. ORNOSKI
nesses, the jury would have heard convincing evidence about
the existence of blood at the crime scene that did not belong
to either DeBaun or Duncan. The jury would have heard how
that evidence points to the existence of an accomplice who
accompanied Duncan to the restaurant that night and who was
injured during the ensuing struggle with DeBaun. The jury
would also have heard that none of the blood found in the
money room belonged to Duncan. Together with the detec-
tive’s testimony that the small room in which the actual kill-
ing took place could likely accommodate only one killer, the
serology evidence that Cheroske should have presented would
strongly support the theory that the accomplice was the actual
and only killer.

   If Cheroske had investigated the blood evidence, he could
have presented the expert testimony of a serologist who
would have testified that Duncan, like the victim, has Type O
blood. That expert could have explained to the jury that
although a great deal of Type O blood was found at the crime
scene, none of it contained detectable subtypes that were con-
sistent with the subtypes in Duncan’s blood.

   Moreover, if Cheroske had investigated the blood evidence,
he would have been prepared to effectively cross-examine
Matheson, the State’s serology expert. At trial, Matheson tes-
tified that contaminants could have produced the test results
in Items No. 5, 8, and 10, which made it easy for the jury to
dismiss the questions raised by those samples. If on cross
examination Cheroske had been informed about serology,
however, he could have asked Matheson how likely it was
that the samples were contaminated. Matheson would have
been forced to respond, as he did in his post-trial declaration,
that the possibility of contamination by other sources was
“not seen very often” and was of “the lowest possibility.”

  Also, if Cheroske had been prepared to cross-examine
Matheson, he could have elicited the testimony about Items
No. 5, 8, and 10 that Matheson gave in his post-trial declara-
                         DUNCAN v. ORNOSKI                         7341
tion. If questioned further, Matheson would have testified that
the stain on Item No. 5 was blood and that it likely contained
blood that was not the same type as Duncan’s blood or the
victim’s blood. Cheroske could have emphasized through
cross-examination that Item No. 5 was taken from a sample
of blood that was large enough to make a partial shoeprint.
Thus, it was unlikely that it could have been deposited earlier
in the day, before the crime, and escaped the notice of the
employees and the cleaning crew.

   With respect to Item No. 8, the stained Handiwipe rag,
Matheson would have testified if questioned further that Item
No. 8 likely contained a mixture of body fluids from someone
with Type O blood and someone with Type AB blood or one
person with Type A and one person with Type B.8 Moreover,
Matheson would have testified that the antigenic activity that
he detected in Item No. 8 came from a human source rather
than contamination because he tested the unstained portion of
the rag and detected no antigenic activity.

   As to Item No. 10, the stain lifted from the floor near the
door between the money room and the bar area, Matheson
would have testified that although the color of the stain was
unusual, it likely came from the body fluid of someone with
Type AB blood. As to that item, Cheroske could also have
introduced into evidence Detective Lewellen’s property
report, which he prepared the day after the stain was lifted,
and in which he described the stain as “red.”

   If Cheroske had cross-examined Matheson effectively, the
jury would have heard Matheson’s final conclusion that the
most likely explanation of the blood evidence “is that there
  8
   The latter scenario is unlikely as it would involve three people shed-
ding blood and staining the Handiwipe rag on the night of the murder, two
of whom could not be Duncan or the victim. There is no suggestion by any
participant in the proceeding that Duncan had two accomplices, nor is
there any other evidence to support such a theory.
7342                   DUNCAN v. ORNOSKI
was human body fluid at the crime scene that came from at
least one person who did not have Type O blood, and that the
body fluid was deposited at the crime scene after the last time
the floor was washed.”

   In addition to effectively cross-examining Matheson,
Cheroske should have had his own serology expert testify at
trial. Such an expert would have given testimony similar to
that of Blake, the serology expert hired by post-conviction
counsel. Like Blake, he could have testified that the “only rea-
sonable and non-speculative interpretation” of the blood evi-
dence and Matheson’s test results was that on the night of the
murder, “someone with Type AB blood was injured, shed
blood on the money room floor, and wiped off some . . . on
the blue and white Handiwipe rag.” That person could not
have been the victim or Duncan, because both of them have
Type O blood.

   Furthermore, a defense serology expert could have clarified
that the possibility that the antigenic activity detected in Items
No. 5, 8, and 10 could have come from non-blood body fluid
was infinitesimally small. According to Blake:

    [T]he AB antigens were on three separate items
    which have blood on them. It would be an extraordi-
    nary and highly unlikely set of coincidences for all
    three items to happen to have saliva or perspiration
    or another bodily fluid on them. Second, the non-
    stained part of Item No. 8, the Handiwipe rag, was
    tested and showed no antigenic activity. This means
    that if blood were not the source of the “A” and “B”
    antigens on Item No. 8, then the further coincidence
    is required that the other body fluid was deposited
    only on the part of the Handiwipe rag that had a
    blood stain on it. Third, as to the red stains from the
    money room floor (Item No. 5 and Item No. 10),
    again the coincidence would be required that the
    non-blood bodily fluid should have been located in
                          DUNCAN v. ORNOSKI                          7343
      two separate places where there was blood on the
      floor, but not on the part of the floor from which the
      control Item No. 4[9] was taken. . . . Although it may
      be possible to concoct a scenario that theoretically
      could account for non-blood fluids on all three items,
      such a scenario would be complex, speculative,
      implausible and improbable.

   If Cheroske had consulted a serology expert and arranged
for Duncan’s blood to be tested, Cheroske would have under-
stood the importance of establishing that the floor had been
mopped the night of the murder. He could have called Mark
Christian and Arnell Jackson, two witnesses who indepen-
dently told police that they observed a custodian mopping the
floor the night of the murder, to testify. Additionally, Fei-
mann, the bartender would have testified that the floors were
mopped every night with soap and water and that the person
responible for mopping the floors was working the night of
the murder. This testimony would have been more than suffi-
cient to convince the jury that the floor had been washed
before the murder occurred so that any blood that was found
at the crime scene must have been shed during the attack.

   Certainly the testimony that Cheroske could and should
have elicited from Christian, Jackson, and Feimann would
have been enough to dispel any doubt that might have been
raised by Goodell, the State’s witness. In her post-trial decla-
ration, Goodell claimed that the night shift was short one util-
ity worker the night of the murder. She also stated that the
mopping was not always thorough. However, Cheroske’s
serology expert could have explained that a thorough mop-
ping was not necessary and that a routine mopping was suffi-
cient to remove antigens from the floor.
  9
   Item No. 4 was a cloth that was used to lift a bloody shoe print from
the money room floor. The shoe print was located approximately four feet
from the shoe print that was used in Item 5. The sample tested positive for
Type O blood and negative for A and B antigenic activity.
7344                  DUNCAN v. ORNOSKI
   Even if the floor was not mopped, or was not mopped thor-
oughly, it is difficult to believe that the blood sample from
which Item No. 5 was made would have been overlooked dur-
ing the day. Item No. 5 was taken from a bloody shoe print.
It is unlikely that a bloody shoe print (or a sufficient amount
of blood to create a shoe print), would have escaped notice
during the day and remained on the money room floor. It is
equally unlikely where there are two bloody shoe prints on the
floor in a small room that one came from blood spilled during
the course of a murder and the other from blood that hap-
pened to be on the floor for several hours prior to the killing
but went unnoticed by those in the room earlier. Moreover, it
is unlikely that an injury that produced enough blood to create
a shoe print would have occurred in the money room. Presum-
ably, any such earlier injury would have occurred in the
kitchen where there were knives. The only plausible explana-
tion for the blood sample is that DeBaun wounded her
attacker and he shed blood on the money room floor.

   The district court found that Duncan did not “present con-
clusive evidence that the floor of the restaurant was mopped
on the night of the murder” or establish that “the cleaning was
thorough enough to remove blood or other bodily fluids from
the floor of the money room.” The burden of proof that the
district court imposed—proof by conclusive evidence—is not
the correct standard. Under Strickland, Duncan has to prove
only that the omitted evidence is sufficient to undermine con-
fidence in the 

. Given that the
floors were customarily washed every night, that there were
two witnesses who saw the floors being washed on the night
of the murder, and that the employee responsible for washing
the floors was working that night, it is reasonable to conclude
that at least one juror would have believed that the floors were
mopped after the restaurant closed and before the murder took
place.

   The question whether the floor was washed is relevant to
Items No. 5 and 10, but not Item No. 8, the blood sample
                      DUNCAN v. ORNOSKI                     7345
taken from the bloody Handiwipe rag, because that sample
was not lifted from the floor. The circumstances under which
the rag was found strongly suggest that the bloody rag was
used after the murder by the assailant who was injured in the
attack. First, the rag was found lying below an open first aid
kit that hung on the wall. The kit was missing supplies, lead-
ing police to speculate that the injured attacker used the rag
to absorb some of his blood and then used the supplies to treat
his wounds. Second, the rag was found lying on top of the
padlock to the key box. As the State explained in closing
argument, the assailant would have had to unlock and remove
the padlock from the key box in order to obtain the key that
opened the floating fund bank. Because the key box was
always kept closed and locked, the padlock must have been
removed by the perpetrators of the robbery. Moreover,
because the rag was found on top of the padlock, it is
extremely unlikely that the blood on the rag belonged to a res-
taurant employee who was injured during the day and left the
bloody rag in the money room, as suggested by the State.
Under that scenario the attacker would have had to slip the
padlock underneath the bloody rag. The far more likely sce-
nario is the one proposed by the detectives—that DeBaun’s
attacker was wounded in the struggle, used the rag and the
first aid supplies from the kit to treat his wounds, and left the
rag lying on top of the padlock that he had previously
removed from the key box in order to open the safe. Clearly,
the actual killer would be the one who was injured during the
struggle so the fact that the blood found on that rag did not
belong to Duncan raises doubt that he killed DeBaun.

   If Cheroske had investigated the blood evidence, tested
Duncan’s blood, and presented the accomplice theory to the
jury, it is likely that at least one juror would have had a rea-
sonable doubt that Duncan was the one who killed DeBaun.

  At trial, Cheroske tried to establish that other individuals
may have had a motive to kill DeBaun, but he never presented
any physical evidence tying anyone else to the crime scene.
7346                       DUNCAN v. ORNOSKI
In fact, all of the physical evidence that was presented at trial
tied Duncan to the crime scene. Additionally, there was no
evidence or argument presented at trial that Duncan had an
accomplice. Thus, given the strong physical evidence tying
Duncan to the scene and the lack of any other explanation for
the murder, it is not surprising that the jury believed that Dun-
can was the actual killer and that he killed DeBaun intention-
ally. The serological evidence introduced during the post-
conviction proceeding, however, raises doubts as to whether
it actually was Duncan who engaged in the physical confron-
tation with DeBaun. The blood evidence shows that someone
other than DeBaun and Duncan shed blood in the money
room on the night of the murder. Given the testimony at trial
that some of DeBaun’s injuries indicate that there was a strug-
gle during the attack and that the size of the money room
made it unlikely that more than one person actually killed
DeBaun or was in the room at the time, the individual who
shed blood the night of the murder was in all likelihood the
killer. Because the blood evidence does not indicate that Dun-
can shed blood, but tends to establish that an accomplice did,
it undermines our confidence in the jury’s finding that Dun-
can was the actual killer. If the jurors had heard the additional
evidence and counsel had presented the accomplice theory,
we believe there is a reasonable probability that one or more
of them would have had a reasonable doubt that Duncan was
guilty of the special circumstance alleged.10
   10
      The district court cited Spivey v. Rocha, 

, 978 (9th Cir.
1999), for the proposition that omitting evidence of third party participa-
tion that does not identify a possible suspect is not reason to grant peti-
tioner relief. However, Spivey involved a state question about the
admissibility of certain evidence in a murder trial. 

In that case,
the defendant claimed in his federal habeas petition that a decision of the
state trial judge to exclude certain evidence constituted constitutional
error. 

case involves a completely different question:
namely, whether counsel was reasonable in failing to investigate available
evidence and how that failure prejudiced the defendant’s case. Only state
evidentiary rulings that render the proceeding so fundamentally unfair as
to violate due process are grounds for federal habeas relief. 

                          DUNCAN v. ORNOSKI                           7347
   The only remaining question is whether once the jury had
a reasonable doubt that Duncan was the killer, a reasonable
doubt would also exist as to whether Duncan intended that his
accomplice kill DeBaun. Once the jury had determined that
Duncan was the actual killer, the jury would also conclude
that he intended the killing. So, none of the evidence pre-
sented at trial speaks directly to this question. However, the
question need not be answered definitively in order for us to
find prejudice. We need only determine whether there is a
reasonable probability that the outcome of the proceeding
would have been different had the additional blood evidence
been presented. See 

. In other
words, we must determine whether we are confident that the
jury, having heard the additional blood evidence that
Cheroske should have presented, would unanimously find
beyond a reasonable doubt that although Duncan was not the
killer, he intended DeBaun’s death. We conclude that there is
a reasonable probability that at least one juror would have
harbored doubt about Duncan’s intent, and that undermines
our confidence in the special circumstance finding.

   The evidence supports the theory that Duncan and his
accomplice intended to rob the restaurant after DeBaun left
for the night and did not expect to encounter her there. The
record indicates that DeBaun worked later than usual on the
night of the murder. If so, she may have caught Duncan and
his accomplice by surprise. Duncan’s accomplice might well
have attacked her without Duncan intending that he do so.
Alternatively, Duncan and his confederate may have known

That is a much higher standard than the one Duncan must meet to show
that Cheroske’s assistance was ineffective. More important, in Spivey, the
only evidence of third party participation was the fact that the victim was
a member of a gang and there was no additional evidence linking another
perpetrator to the murder. 

In this case, by contrast, there is
forensic evidence that places a third party at the scene of the crime on the
night of the murder and demonstrates that the third party was injured dur-
ing the commission of the crime.
7348                  DUNCAN v. ORNOSKI
that DeBaun was still in the money room and worn masks to
disguise their identities when they entered the restaurant.
Although Duncan may have intended only to steal the money
and leave DeBaun alive, his accomplice may have decided to
kill her without Duncan’s consent, when he unexpectedly
found her still there or when she put up an unexpectedly
strong struggle instead of simply turning over the money.
Either of these scenarios would be consistent with the evi-
dence presented at trial and at the state post-conviction pro-
ceedings. More important, as the district court acknowledged,
if an accomplice killed DeBaun, it would be difficult to estab-
lish from the evidence before the jury that Duncan intended
that he do so, especially as the State sought to prove that there
was only one participant in the robbery and offered no evi-
dence that Duncan intended the accomplice who actually
killed DeBaun to murder her.

   None of the arguments that the district court or the State
offer in opposition to the new blood evidence forecloses the
reasonable possibility that at least one juror would have found
that Duncan did not intend to kill DeBaun. The fact that
DeBaun was Duncan’s supervisor and thus surely would have
recognized him does not conclusively prove that he intended
her death. Duncan is not a sophisticated or strategic criminal
as demonstrated by the second theft in which he stole $1700
from the supervisor’s bank and revealed his identity as the
thief by also stealing $300 from his own cashier bank. Nor
does the fact that two knives were used in the attack prove
that Duncan assisted in the attack, especially considering the
detective’s testimony that the money room was too small for
two attackers. In fact, the jury convicted Duncan on the theory
urged by the State that one attacker wielded two knives
against DeBaun.

   The district court found that Cheroske’s failure to present
the additional blood evidence did not prejudice Duncan, but
the court applied the incorrect legal standard. Throughout its
order, the district court found that petitioner did not present
                      DUNCAN v. ORNOSKI                   7349
“conclusive evidence” of his arguments or “establish” crucial
facts. (E.g., floor mopping, two attackers, AB blood deposited
on the night of the murder). The district court also ruled that
the State could have offered “persuasive” evidence rebutting
petitioner’s arguments. (E.g., AB blood came from accom-
plice, explanation of two knives, intent to kill). The “conclu-
sive evidence” standard articulated by the district court, as
well as the court’s speculation about what the prosecution
“could have” offered, hold petitioner to a standard well above
the “reasonable probability” required by Strickland. Even the
lesser “more likely than not” or “preponderance” require-
ments were “explicitly rejected in Strickland.” Sand

. A requirement that petitioner establish his con-
tentions “conclusively” conflicts directly with Strickland. In
fact, in order to prevail on this appeal, Duncan need show
only that the omitted evidence is sufficient to undermine con-
fidence in the outcome, and he has done so.

   Not only was Cheroske’s deficient performance with regard
to the serology expert and the blood evidence “far less helpful
than a competent presentation would have been,” but it was
also “probably actually harmful” to Duncan’s case. 

. In his closing argument, the prosecutor criti-
cized Cheroske for not disclosing Duncan’s blood type and
implied that his failure to do so was indicative of guilt. He
asked the jury: “don’t you think that the defense would have
taken the blood of this defendant and brought it into court
here with their own expert and had them testify to it if they
thought there was the slightest chance to be able to prove
that?” The prosecutor’s argument suggested that there existed
additional inculpatory evidence that was never presented to
the jury—a claim that certainly undermined Duncan’s
defense. See 

(finding prejudice when the prosecutor
capitalized on defense counsel’s deficient performance and
used it to his advantage in closing arguments).

  [16] In sum, Cheroske’s failure to properly investigate the
blood evidence, test Duncan’s blood, and present evidence
7350                  DUNCAN v. ORNOSKI
that someone other than Duncan and the victim shed blood in
the money room on the night of the murder was unreasonable,
especially considering that the blood evidence was the only
physical evidence that had not been linked to Duncan at the
time of the trial. The evidence that Cheroske failed to present
would have been highly significant because it would have
suggested that Duncan had an accomplice and that the accom-
plice was likely the actual killer. Under the State’s own the-
ory, the small money room likely would have accommodated
only one killer. Given the blood found at the crime scene that
did not belong to the victim or to Duncan and that was likely
shed in the course of the attack, it appears probable that Dun-
can was not in the money room during the murder.

   [17] Duncan has undermined our confidence in the jury’s
special circumstance finding that he intended to kill and did
kill DeBaun because there is a reasonable probability that if
the jury had heard the additional blood evidence, at least one
juror would have had a reasonable doubt as to the truth of that
allegation. For the reasons set forth above, we hold that with
respect to the special circumstance finding, the district court
erred in concluding that Duncan suffered no prejudice from
Cheroske’s failure to investigate and present testimony
regarding the critical blood evidence.

                         Conclusion

   We affirm the district court’s denial of habeas relief as to
Duncan’s conviction for first-degree felony murder for the
reasons set forth above and in the memorandum disposition
filed concurrently herewith. We also hold that the district
court erred in concluding that Duncan did not receive ineffec-
tive assistance of counsel at the special circumstance phase of
his trial. Accordingly, the special circumstance finding must
be vacated, and the sentence set aside. The district court shall
issue a writ directing the State to institute within 90 days any
proceedings necessary to permit a jury to make new special
                     DUNCAN v. ORNOSKI                  7351
circumstance findings or, alternatively, to impose a sentence
of 25 years to life in prison.

  AFFIRMED in part, REVERSED in part, and
REMANDED with instructions to issue a writ of habeas cor-
pus.