Facial composite research has generally focused on the investigative utility of composites--using composites to find suspects. However, almost no work has examined the diagnostic utility of facial composites--the extent to which composites can be used as evidence against a suspect. For example, detectives and jurors may use the perceived similarity of a suspect to a composite as evidence to determine the likelihood of a suspect's guilt. However, research in social cognition and models of cognitive coherence suggest that these similarity judgments may be biased by evaluators' preexisting beliefs of guilt. Two studies examined how preexisting beliefs of guilt influence similarity ratings between a suspect and a facial composite. Study 1 (n = 93) demonstrated that mock-investigators' beliefs in a suspect's guilt inflated their subsequent similarity ratings. Study 2 (n = 49) demonstrated that mock-jurors' beliefs in a defendant's guilt predicted their similarity ratings. These findings highlight a problem of using facial composites as evidence against a suspect, and demonstrate the malleability of similarity judgments. (Contains 3 tables and 3 figures.)