Abstract: Classical trust model of P2P networks calculates the global trust value by iteration of local trust value. Every transaction will cause iteration throughout the whole networks resulting in computational complexity, huge communication traffic. These also face collusion attack, smear attack, sleeping attack and so on that caused by sparse transaction data and inaccurate computing result. To ensure the density of transaction data and the accuracy of computing result, a novel P2P global Probability and Statistics based trust (PStrust) model is presented. The history records of transaction are used to figure out the trust value of every peer by methods of the maximum likelihood estimation and hypothesis testing. Every peer trades with the peer with high credibility. Mathematical analysis and simulation show PStrust can resist attacks of malicious peers and improve the successful download rate of the whole P2P system compared with traditional model Eigentrust.