How do we make inferences about the moral character of others? Here we review recent work on the cognitive mechanisms of moral inference and impression updating. We show that moral inference follows basic principles of Bayesian inference, but also departs from the standard Bayesian model in ways that may facilitate the maintenance of social relationships. Moral inference is not only sensitive to whether people make moral decisions, but also to features of decisions that reveal their suitability as a relational partner. Together these findings suggest that moral inference follows a relational logic: people form and update moral impressions in ways that are responsive to the demands of ongoing social relationships and particular social roles. We discuss implications of these findings for theories of moral cognition and identify new directions for research on human morality and person perception.