Responding appropriately to others’ facial expressions is key to successful social functioning and, depending on the identity of your interactive partner, it might save you from harm. Despite the large body of work on face perception and spontaneous responses to static faces, little is known about responses to faces in dynamic, naturalistic situations, and no study has investigated how people’s goal directed responses to faces are influenced by learning during dyadic interactions. Addressing this question, we developed a novel method based on online integration of electromyography (EMG) signals from the participants’ face (corrugator supercilii and zygomaticus major). In this experiment (N = 58), participants’ EMG signals were recorded during their interaction with dynamic faces that either displayed happy or angry facial expressions. The participants’ task was to either reciprocate (congruent) or respond in the opposite manner (incongruent). Incorrect facial expressions were punished with a mild electrical shock to the participant’s wrist. Our results validated our new method, showing that participants learned to optimize their facial behavior over time, and replicated earlier findings of faster and more accurate responses in congruent vs. incongruent conditions. Moreover, participants performed better on trials when confronted with smiling, as compared to frowning, faces, suggesting it might be easier to adapt facial responses to positively associated expressions. Our results introduce a new method to study learning and decision-making in dynamic interactive social situations, in which there is a need to gradually adapt facial mimicry to both social and non-social reinforcements.