The paper examines rational cooperation in the finitely repeated prisoner's dilemma through experimental evidence. It tests the sequential equilibrium reputation hypothesis, which suggests that players may cooperate early in the game to build a reputation, even if they eventually defect. The study involves four conditions: Partners, Strangers, Computer50, and Computero. In the Partners condition, subjects play with new partners each round, while in the Strangers condition, they face new opponents each round. In the Computer50 condition, there is a 50% chance of playing a computer partner, and in the Computero condition, the chance is 0.1%. The results show that subjects in the Partners and Computer50 conditions are more cooperative than those in the Strangers condition. Cooperation is highest in the early rounds and declines near the end, consistent with the reputation-building hypothesis. The Computer50 condition shows significantly more cooperation than the Partners condition, suggesting that the belief in an altruistic opponent increases cooperation. The Computero condition, where the tit-for-tat strategy is common knowledge, shows similar cooperation levels to the Partners condition, indicating that common knowledge alone may not be sufficient for cooperation. The study also finds evidence of altruism among subjects, with some individuals cooperating consistently even when there was no opportunity to build a reputation. This suggests that a significant portion of the population may have altruistic tendencies. The results support the sequential equilibrium reputation model, showing that subjects are willing to build reputations for altruism and that this behavior is consistent with the predictions of the model. The paper concludes that the sequential equilibrium reputation hypothesis is a good predictive model of cooperative behavior in the finitely repeated prisoner's dilemma, and that there is a significant number of altruistic types in the population.The paper examines rational cooperation in the finitely repeated prisoner's dilemma through experimental evidence. It tests the sequential equilibrium reputation hypothesis, which suggests that players may cooperate early in the game to build a reputation, even if they eventually defect. The study involves four conditions: Partners, Strangers, Computer50, and Computero. In the Partners condition, subjects play with new partners each round, while in the Strangers condition, they face new opponents each round. In the Computer50 condition, there is a 50% chance of playing a computer partner, and in the Computero condition, the chance is 0.1%. The results show that subjects in the Partners and Computer50 conditions are more cooperative than those in the Strangers condition. Cooperation is highest in the early rounds and declines near the end, consistent with the reputation-building hypothesis. The Computer50 condition shows significantly more cooperation than the Partners condition, suggesting that the belief in an altruistic opponent increases cooperation. The Computero condition, where the tit-for-tat strategy is common knowledge, shows similar cooperation levels to the Partners condition, indicating that common knowledge alone may not be sufficient for cooperation. The study also finds evidence of altruism among subjects, with some individuals cooperating consistently even when there was no opportunity to build a reputation. This suggests that a significant portion of the population may have altruistic tendencies. The results support the sequential equilibrium reputation model, showing that subjects are willing to build reputations for altruism and that this behavior is consistent with the predictions of the model. The paper concludes that the sequential equilibrium reputation hypothesis is a good predictive model of cooperative behavior in the finitely repeated prisoner's dilemma, and that there is a significant number of altruistic types in the population.