Experiment participants in a social dilemma game choose cooperation over defection, even though neither is more beneficial. High levels of cooperation cannot be explained by favorable labels for actions, collusion, k-level reasoning, quantal response behavior, or misplaced optimism about others' actions, but can be rationalized by the Charness and Rabin (2002) preference model. However, cooperation rates fall with changes in payoffs, which cannot be explained by the standard formulation; to account for these results, we introduce a generalization of the model.