People can form opinions of other individuals based on information about their good or bad behavior. The present study investigated whether this affective learning might depend on memory links formed between initially neutral people and valenced information. First, participants viewed neutral faces paired with sentences describing prosocial or antisocial behaviors. Second, memory suppression manipulations with the potential to aid in the forgetting of valenced information were administered. Using the Think/No think paradigm, the effectiveness of four different suppression instructions was compared: Unguided Suppression, Guided Suppression, Distraction, and Thought Substitution. Overall, all the tasks appreciably reduced affective learning based on prosocial information, but only the Guided Suppression and Thought Substitution tasks reduced affective learning based on antisocial information. These results suggest that weakening the putative memory link between initially neutral people and valenced information can decrease the effect of learned associations on the evaluation of other people. We interpreted this as indicative that social affective learning may rely on declarative memories.