We present a version of the experience-weighted attraction (EWA) reinforcement learning model that integrates norm conformity into its utility function that we call “EWA+Norms.” We compare the behavior of this hybrid model to the standard EWA when applied to a step-level public good game in which groups of agents must reach a threshold level of cooperation to avoid the risk of catastrophe. We find that standard EWA is not sufficient for generating cooperation, but that EWA+Norms is. We aim to compare simulation results with human behavior in large-scale experiments that we are currently running.