Knowing when to trust others is an important social skill, but recent findings suggest that humans struggle with this dilemma—trusting strangers more than they should. Although trust decisions often do not meet the standards of rationality, they appear to be boundedly rational. We present a model of heuristic trust, according to which people focus on their own potential outcomes (what may be gained or lost from trusting), but neglect the probabilities of those outcomes occurring. We examine how trustors form expectations of reciprocity, and how those expectations relate to optimal trust decisions: some previous research suggests that people underestimate the probability of reciprocity and, relative to their subjective expectations, trust strangers too much. In contrast, our heuristic model allows for fine-grained predictions of when people trust too much and when they trust too little. The accuracy of trust depends on the selection and use of available cues; errors occur when trustors neglect valid, but difficult to process, cues and overemphasize salient cues lacking validity.