Citing classic psychological research and a smattering of recent studies, Kassin, Dror, and Kukucka (2013) proposed the operation of a forensic confirmation bias, whereby preexisting expectations guide the evaluation of forensic evidence in a self-verifying manner. In a series of studies, we tested the hypothesis that knowing that a defendant had confessed would taint people's evaluations of handwriting evidence relative to those not so informed. In Study 1, participants who read a case summary in which the defendant had previously confessed were more likely to erroneously conclude that handwriting samples from the defendant and perpetrator were authored by the same person, and were more likely to judge the defendant guilty, compared with those in a no-confession control group. Study 2 replicated and extended these findings using a within-subjects design in which participants rated the same samples both before and after reading a case summary. These findings underscore recent critiques of the forensic sciences as subject to bias, and suggest the value of insulating forensic examiners from contextual information.