Abstract: Considering the shortage of traditional modeling and simulation, digital simulation modeling of automatic train supervision （ATS） based on agent is discussed and studied, in order to meet the urgent need of urban transit systems research in China. System framework and semantically based grammar of multi-agent system are studied and designed. Frame models of six kinds of agents are built up. Their function and control flow are constituted. A visual ATS simulation system based on agent is developed with VC⁺⁺6.0 using time-step method and activity scanning strategy. Based on the inherent advantages of agent, the difficulty of system design is greatly reduced. The ATS simulation system based on multi-agent possesses its features in reusability and expansibility.