Under what conditions do social preferences, such as altruism or a concern for fair outcomes, generate efficient trade? I analyze theoretically a simple bilateral exchange game: Each player sequentially takes an action that reduces his own material payoff but increases the other player’s. Each player’s preferences may depend on both his/her own material payoff and the other player’s. I identify necessary conditions and sufficient conditions on the players’ preferences for the outcome of their interaction to be Pareto efficient. The results have implications for interpreting the rotten kid theorem, gift exchange in the laboratory, and gift exchange in the field.