Long-term free-riding behavior of Peer to Peer(P2P) systems makes the system gradually lose its original distributed shared characteristics. In order to prevent selfish behaviors, a new evaluation model for P2P systems is proposed based on the characteristics of game theory suppression model and incentive-based node behavior model. The reputation value which reflects the period performance of users is adopted to characterize user-behavior, and some confinement conditions are set to prevent generating selfish nodes. The various kinds of incentive mechanisms in P2P systems are mathematically modeled and fitted with Auto-Regressive(AR) model. The simulation results based on this model show that the behavior of the system evaluation changes periodically; and the satisfaction is stabilized to a smooth number, 30%.