To estimate evacuation capacity of subway stations under emergency conditions, a bilevel programming model of evacuee equilibrium is built by considering travel time on walking facilities with various congestion degrees. The upper-level model ensures the maximum utilization of facilities, whereas the lower level model is to minimize the evacuation time by determining how to guide evacuees to arrive at safety zones. An improved particle swarm optimization algorithm is designed to solve the model. To validate the model and algorithm, by taking capacity estimation of Fuxingmen Station of Beijing Subway as an example, evacuation process is simulated to get the optimal routes and proper estimation. Evacuation passenger flow per unit time and network crowdedness are selected as indexes to estimate the capacity of station under emergency condition. The results show that the capacity of Fuxingmen Station is 1,071 persons per minute in the adverse situation.