Abstract : According to an in? uential view, one function of mirror neurons (MNs), ? rst discovered in the brain of monkeys, is to underlie third-person mindreading. This view relies on two assumptions: the activity of MNs in an observer ' s brain matches (simulates or resonates with) that of MNs in an agent's brain and this resonance process retrodictively generates a representation of the agent ' s intention from a perception of her movement. In this paper, I criticize both assumptions and I argue instead that the activity of MNs in an observer ' s brain is enhanced by a prior representation of the agent ' s intention and that their task is to predictively compute the best motor command suitable to satisfy the agent ' s intention.