Bi-level programming for designing taxi quantity was established based on the dynamic equilibrium relationship between the fleet size and taxi demand. The upper model employed a double-objective function to minimize the passenger waiting time at peak hours and the taxi unloaded ratio at off-peak hours; the lower model provided a method to decide the demand when the quantity of taxi was known. Finally, the proposed solution was calculated based on multi-object genetic algorithm and linear iteration, and validated using the taxi statistics in Dalian. The results show that, the taxi quantity is not adequate and leads to longer passenger waiting time. It is consistent with the actual conditions in Dalian. The results also show that the expanding of capacity scale can attract a large number of taxi demands when there is a large lack of capacity scale. However, the demand for capacity growth is no longer sensitive when demand and capacity scale are close.