Across six studies (N = 1988 US residents and 81 traditional people of Papua), participants judged agents acting in sacrificial moral dilemmas. Utilitarian agents, described as opting to sacrifice a single individual for the greater good, were perceived as less predictable and less moral than deontological agents whose inaction resulted in five people being harmed. These effects generalize to a non-Western sample of the Dani people, a traditional indigenous society of Papua, and persist when controlling for homophily and notions of behavioral typicality. Notably, deontological agents are no longer morally preferred when the actions of utilitarian agents are made to seem more predictable. Lastly, we find that peoples’ lay theory of predictability is flexible and multi-faceted, but nevertheless understood and used holistically in assessing the moral character of others. On the basis of our findings, we propose that assessments of predictability play an important role when judging the morality of others.