To reflect the influence of uncertain demand to bus frequencies, transit network robustness performance was adopted to optimize the bus frequecies. A bi-level programming model was presented in the process. After the analysis of demand variation propagation in the passenger assignment process, the upper-level model was put forward by the analytical method. There were two items in this model, the former was transit network system benefit consisted of passengers’ costs and operating costs, and the latter was transit network robustness performance. The objective of lower-level was to assign transit trips to bus route network based on optimal strategy. The adopted genetic algorithm to the model was illustrated with a test network and a city network of Liupanshui. It indicated that the transit network robustness performance could improve the reliability of bus frequencies.