In dynamic networks such as MANETs and p2p, trust is an important ingredient. Traditional trust models based on reputation can restrain some ordinary behaviors of malicious nodes and avoid them abusing of resources. However, it suffers a lot from dishonest feedbacks and strategically cheating behaviors of malicious peers. According to the behavior similarity rule of the human society and the definition of the mathematic probability, a behavior similarity inspection module (BSIM) is presented. Users can set inspection tightness parameter along with the network actual circumstances appropriately. Experiment results show that our design can make the trust model more robust and can achieve better trade performance in dynamic networks.