People’s emotional states often depend on the emotions of others. Consequently, to predict their own responses to social interactions (i.e., affective forecasts), we contend that people predict the emotional states of others (i.e., empathic forecasts). We propose that empathic forecasts are vulnerable to stereotype biases and demonstrate that stereotypes about the different emotional experiences of race (Experiment 1) and sex groups (Experiment 2) bias empathic forecasts. Path modeling in both studies demonstrates that stereotype-biased empathic forecasts regarding how a target individual will feel during a social interaction are associated with participants’ affective forecasts of how they will feel during that interaction with the target person. These affective forecasts, in turn, predict behavioral intentions for the social interaction before it even begins. Stereotypes can therefore indirectly bias affective forecasts by first influencing the empathic forecasts that partly constitute them. In turn, these potentially biased affective forecasts determine social behaviors.