Two groups of subjects were induced to form opposing beliefs about the relationship between two social variables, preference for risk and performance in firefighter jobs. Specifically, one group created a causal explanation for a positive relationship (high risk preference related to high performance and low risk preference related to low performance), whereas the other created an explanation for the opposite, negative relationship. Subsequent assessment of subjects' own beliefs revealed a tendency to believe in whichever relationship they had explained. Both groups then examined and evaluated inconclusive "case histories" (half supporting a positive relationship and half supporting a negative relationship). A new assessment of subjects' beliefs showed that this new evidence had little if any effect; subjects' evaluations of the new evidence gave no indication that the groups had evaluated the evidence differently. Why beliefs that are newly formed, and based on no valid evidence survive non-supportive evidence is unclear, but a few ideas are briefly discussed and possible directions for future research are proposed.