Because costly punishment is not credible, subgame perfection suggests that punishment will not deter free riding, regardless of the size or structure of groups. However, experiments show that people will punish free riders, even at considerable cost. To examine the implications of agents who punish, we simulate an environment populated with behavioral strategies seen in the lab and use the simulation to develop hypotheses about why group size should matter when punishment is allowed. We test these hypotheses experimentally and examine whether the effect of group size is purely due to the number of group members or if information about other group members is what is important. We find that large groups contribute at rates no lower than small groups because punishment does not fall appreciably in large groups. However, hindrances to monitoring do reduce the provision of the public good.