Mirror neuron's common coding of action perception and action production raises exciting possibilities for understanding imitation's neural basis. Thus far, however, theorizing on this score has been hampered by oversimplification of imitation's complex cognitive reality. This chapter provides a starting point for addressing this problem, synthesizing a broad range of behavioral research into a theoretically unified conception of human imitation. Particular attention is paid to the problem of reconciling selective imitation, in which children rationally omit unnecessary aspects of an adult's behavior, with overimitation, in which they seem to do exactly the opposite. The apparent contradiction is resolved by refining the notion of selective imitation to its underlying computational basis, and showing that this computational substrate in fact predicts observed patterns of overimitation. Selective imitation and overimitation are thus re-conceptualized as different sides of the same coin—two rational social learning strategies whose superficial opposition conceals a deeper mechanistic and conceptual commonality.