In decentralized peer-to-peer file-sharing networks, due to the anonymous and self-organization nature of peers, they have to manage the risk involved with the transactions without prior knowledge about each other’s reputation. SWRTrust, a global trust model, is proposed to quantify and to evaluate the trustworthiness of peers, which includes a mathematical description and a distributed implementation. In SWRTrust, each peer is assigned a unique global trust value, computed by aggregating similarity-weighted recommendations of the peers who have interacted with it. Previous global trust models are based on the assumption that the peers with high trust value will give the honest recommendation. This paper argues that this assumption may not hold in all cases. Theoretical analyses and experimental results show that SWRTrust is still robust under more general conditions where malicious peers cooperate in an attempt to deliberately subvert the system, converges more quickly, and decreases the number of inauthentic files downloaded more effectively than the previous models.