Identifying casual peer influence is a long-standing challenge to social scientists. Using data from a natural experiment of randomly-assigned college roommates (N = 2,059), which removes the threat of friend selection, we investigate peer effects on aggressive behavior, smoking, and concurrent sexual partnering. The findings suggest that the magnitude and direction of peer influence depend on predisposition, gender, and the nature of the behavior. Peer effects on individuals predisposed toward a given behavior tend to be larger than peer effects on individuals without such a predisposition. We find that the influence of roommates on aggressive behavior is more pronounced among male students than among female students; roommate effects on smoking are negative among female students and male students who did not smoke before college. For concurrent sexual partnering, a highly private behavior, we find no evidence of peer effects.