Free-riders and fake files are two important problems in P2P file sharing systems. Previous works have always used incentive mechanisms and trust mechanisms to ad- dress them respectively. In real systems however, a trust mechanism without incentive would face lack of users' enthusiasm and thus cause sparse relationship of direct trust while an incentive mechanism without trust could induce users' bad behavior. A novel reputation system is proposed in this paper that combines trust and incentive mechanisms. It uses files' vote and retention time, down- load volume and users' rank to construct a more extensive direct trust relationship and calculates a user's reputa- tion with a multi-trust algorithm. It can identify fake files and provide service differentiation with reputation. Im- plementation and some security consideration in DHT are also discussed.