Abstract: Previous work combining functional neuroimaging with models of reward learning has identified neural correlates of important computations underlying self-interested choice, including expected value and prediction errors. However in many social contexts, decisions are not solely self-interested. Rather, decisions are influenced by both self- and other-regarding preferences. In two studies, we explore how self- and other-regarding preferences are expressed as individuals make decisions for oneself and others. In the first task, we identify neural computations of expected value within a moral hazard task, in which the interests of an individual vary those of other social agents. In a second task, we identify separable learning signals that track and update value for outcomes accrued to oneself or others. These data extend our understanding of value computations made within social contexts, and have implications for neural structures previously associated with ‘mentalizing’ and ‘theory-of-mind’.