Train traffic control system is the most important system of the intelligent railway transportation systems and guarantees the train's safty & punctuality. To solve the problem of train traffic control, this paper studied the application of MAS in the train traffic control by using theoretical method. Firstly, the agent's living enviorments were divided into two parts, dynamic environment & static enviorment, which were described respectively by using the first-order predicate. The formal description for static environment included the relation among stations, between arriving routes and departing routes, between routes and signal lamps, and train operation plan etc. which did not change with time. The formal description for dynamic environments included train dynamic attributes, stations dynamic information, railway line, natural environment, and agent social attribute ect. which changed with time. The reasoning rules were presented after the formal description. Secondly, according to the real situation, this paper studied the agent system architecture and presented the two ways four layers system architecture for railway bureau agent BA (Bureau Agent) and reactive system architecture for station agent SA (Station Agent). In addition, the working principle of BA & SA was expounded in detail in this paper. Thirdly, through analyzing the human social organization structure, we gave the organization structure of MAS and the social rules that each agent must obey. Fourthly, the local planning method was studied after giving the agent system architecture, and the GA_based train operation adjustment model and the conflict-detecting algorithm were presented. Fifthly, according to the situation of train operation adjustment, the negotiation & cooperation method was studied and a new method was presented through which MAS could form the global goal. In the end, the communication method among