In mobile ad-hoc networks (MANETs), there are selfish nodes, which drop the received data packets, because nodes have limits of their resources. Secure routing protocols are developed as one of security mechanisms. In secure routing, each node can detect selfish nodes and decide secure paths by consulting trust values. Here, trust values express reliability for nodes or paths. In these existing routing protocols, all nodes are assumed to keep the same behavior in order to obtain the stable trust values. However, there are selfish nodes with behavior changes because resources for these mobile nodes decrease as time passes. Therefore, as it is hard to calculate accurate trusts in nodes with behavior change in real time, there are two problems. One is to make a mistake in judgment of selfish nodes and the other is not to be able to select the secure paths. In this paper, we propose TEAB, which is a trust evaluation method adapted to node behavior. We introduce the adaptive time window which is the time period for evaluating a trust value. Each node uses a variable time window for each neighbor and reduces the undesirable effects of past behaviors. In this way, each node calculates trust values correctly in the environment including nodes with behavior change. Moreover, in secure routing protocols using TEAB, each node can select the secure communication path composed of only cooperative nodes. The results of simulation show that each node can detect the selfish nodes correctly and start to communicate through the secure path by using TEAB.