Abstract: It is of great importance to introduce trust when solving the collaboration and interaction problem in multi agent system （MAS） environment because like the human society, there are so many indeterminate factors in the MAS environment. Generally speaking, trust derives from direct trust and reputation, and the reputation system is the mechanism to support trust evaluation. The reputation system can be studied at both individual and system level, and more attention should be paid to the former. Whether cognitive view or numerical view is used to represent trust, the trust value evaluated by the reputation system should be accurate, adaptive, and so on. The reputation system can adopt concentrated, distributed or mixture architectures and they all demand the way to represent, propagate and aggregate the trust information. Up to now, the ReGreT and FIRE are both successful among the distributed architecture. As a hot, basic issue in the reputation system, inaccuracy information problem is a pressing topic, which requires to be explored further. Another limitation is that no reputation model test platform receives public recognition, even though the Agent Reputation and Trust （ART） Testbed Project has gained preliminary achievements. Based on the above introduction and review, it is proposed that more work should be done to construct a new model and mechanism, improve the existing models, and develop a new test platform.