To evaluate an individuals' moral character the intentions behind their actions must be discriminated from the actual outcome. How this is achieved remains unclear. We developed a novel paradigm that dissociates the perception of intentions from outcome. Participants separately predicted and received feedback on agents' intentions and outcomes. They then made moral evaluations of the agent. Four independent experiments (120 participants in total) demonstrated that intentions and outcomes mutually influenced each other during the learning process before integrating them for subsequent moral evaluation. Computational modeling further revealed that intentions biased the predictions of outcomes, while outcomes directly modified the beliefs about intentions. Moreover, participants considered both intention and outcome when making moral evaluations, but placed greater weight on intention, regardless of sampling bias and presentation order. These findings offer new insights concerning how individuals process intention and outcome when learning about others' moral character.