With changes in network topology and wireless conflicts, the evidence of trust sampling space may be incomplete or unreliable, which will thus affect the trust evaluation validity. In consideration of inaccurate evaluation in the traditional reputation management mechanism and complexity of the malicious nodes, the Dempster-Shafer (D-S) evidence theory is used to establish the trust evaluation model that can deal with both trust randomicity and subjective uncertainty without prior distribution in evaluation. It also has the recommendation trust filtering mechanism which can monitor the node’s behavior at real time, detect and isolate malicious node in time. Theoretical analysis and simulation results show that the proposed model can provide an effective method for trust uncertainty expression and processing, and can effectively resist the attack and reduce the impact of malicious nodes. All in all, our proposed model can promote the detection accuracy ratio and enhance the packets’ forwarding services.