Utilizing peer-to-peer (P2P) overlay networks to organize resources is a hot topic recently. How to assure the dependability of the resource selection in the overlay is a crucial problem. In overlay networks, self-organization is handled through protocols for node arrival and departure. Topology revolution based on physical location can improve the performance of P2P networks, but it ignores the problem of malicious nodes in the networks, which will greatly affect the resource providing ability of the whole P2P network; while topology revolution based on interest could improve the efficiency of resource sharing and searching in the networks, but it ignores the actual ability of resource providing and the reliability of nodes as well. In this paper a trust mechanism-based self-organization resource topology is developed and Bayesian method is used to evaluate the trust value of nodes based on their behavior. By updating the trust-based connection between each node, a novel self-organization topology is built on the top of the physical topology. Simulation experiments show that the proposed model not only increases the resource find ratio, but also improves the interaction performance of the entire network. In addition, this kind of model can cluster nodes near the trustworthy ones in the networks which have stronger ability of resource providing, and assure the reliability of resource sharing and selection.