The presumption of innocence is not only a bedrock principle of American law, but also a fundamental human right. The psychological underpinnings of this presumption, however, are not well understood. To make progress, one important task is to explain how adults and children infer the goals and intentional structure of complex actions, especially when a single action has more than one salient effect. Many theories of moral judgment have either ignored this intention inference problem or have simply assumed a particular solution without empirical support. We propose that this problem may be solved by appealing to domain-specific prior knowledge that is either built-up over the probability of prior intentions or built-in as part of core cognition. We further propose a specific solution to this problem in the moral domain: a good intention prior, which entails a rebuttable presumption that if an action has both good and bad effects, the actor intends the good effects and not the bad effects. Finally, in a series of novel experiments we provide the first empirical support – from both adults and preschool children – for the existence of this good intention prior.