To solve the problem of load unbalance among servers in large network games, a load balancing strategy based on Dempster/Shafer (D-S) evidence theory was proposed. The multiple factors which influenced the servers were taken as parameters. Firstly, according to D-S evidence theory, the multiple factors affecting the performance of the server were used as the criteria, the dynamic weight was computed by comparing the historic data with the threshold, and then the basic belief function was set up according to the relationship between the dynamic weight and original reliability. After that, the belief functions corresponding to different criteria was calculated, and the calculation results were merged by the rules of evidence synthesis. Lastly, whether the server was overloaded or not was evaluated by the analysis of aforementioned results. Simulation results show that compared with the dynamic load balancing algorithm based on negative feedback, the proposed algorithm is more accurate and more realistic; the running time of the proposed algorithm is obviously less than that of the dynamic load balancing algorithm based on negative feedback and the weighted loop algorithm. Analysis indicates that the proposed algorithm can effectively reduce the delay of running judgement and make a quick deduction for the server load according to the historical parameters, and the dicision results are more reliable and more consistent with the actual situation.