In reputation systems, the anonymous evaluation mechanisms introduced for preserving privacy of honest feedback providers brings about the difficulty in identifying slandering, ballot stuffing and Sybil attacks. A trust model which protects honest feedback providers and identifies and traces the malicious peer was proposed in this paper to deal with this problem. Peers in this trust model use a verifiable random function to generate tags, so as to anonymously evaluate the transaction objects and hide the true identity of the transaction process. In this model, the Bayesian filtering algorithm was introduced to identify malicious tags; when the tags exceed the threshold malicious number, the trust model can automatically expose true identities and track all of the providing feedbacks based on the verifiable secret sharing mechanism. The simulation results show that the proposed trust model can efficiently resist attacks of anonymous malicious peers and evidently improve the accuracy of trust accumulated value compared with two existing trust models.