Abstract: Existing models suggest that reciprocity is unlikely to evolve in large groups as a result of natural selection. In these models, reciprocators punish noncooperation by with-holding future cooperation, and thus also penalize other cooperators in the group. Here, we analyze a model in which the response is some form of punishment that is directed solely at noncooperators. We refer to such alternative forms of punishment as retribution. We show that cooperation enforced by retribution can lead to the evolution of cooperation in two qualitatively different ways. (1) If benefits of cooperation to an individual are greater than the costs to a single individual of coercing the other n − 1 individuals to cooperate, then strategies which cooperate and punish noncooperators, strategies which cooperate only if punished, and, sometimes, strategies which cooperate but do not punish will coexist in the long run. (2) If the costs of being punished are large enough, moralistic strategies which cooperate, punish noncooperators, and punish those who do not punish noncooperators can be evolutionarily stable. We also show, however, that moralistic strategies can cause any individually costly behavior to be evolutionarily stable, whether or not it creates a group benefit.