The present study tests the hypothesis that behavioral information diagnostic of an out-group's traits biases the expected facial appearance of out-group members toward having facial features corresponding with the inferred traits. Participants formed a stereotype about a novel group based on random exemplar faces, presented alongside descriptions of their behavior. The behavioral information was manipulated to reflect either trustworthy or criminal traits, whereas the stimulus faces did not reflect any traits. Afterwards, participants' expected facial appearance of group members was assessed using a reverse correlation task. Independent judges rated the resulting visualized expectations as more criminal in the criminal behavioral information condition than in the trustworthy behavioral information condition. The current work establishes a causal link between behavioral information and expected out-group faces where previously only correlations had been observed. Copyright © 2012 John Wiley & Sons, Ltd.