Trust research is an important issue in P2P networks. Current trust models are lack of flexibility in dynamic environment. This paper presents a fine-grained trust computation model- FineTrust. The main contribution including: offering Gauss-bar function to evaluate the similarity of service satisfaction; considering the reliability of information from different people; setting some parameters in computing trust, such as recommendation consistent, recommendation accuracy, and recommendation error, which make the result of trust evaluation more natural, and so on. Theoretic analyses and simulations show that, the proposed trust model is more precise and more robust.