Abstract: Establishing a trust system, which considers energy efficiency and a trust metric aware both subjectivity and objectivity in the Internet of Things, is a powerful technique to defend against malicious attacks and improve the performance of network. A trust system based on behavior detection was proposed, which takes direct trust, recommended trust as well as history statistical trust into trust evaluation periodically and in communication. Recommended trust and history statistical trust were calculated by evidence combination and Bayes respectively. Simulation results show that nodes occur abnormal behavior could be quickly detected by the proposed trust system and the energy consumption of transmitting recommended trust was saved greatly.