The Nansi Lake is the largest freshwater lake of northern China and consists of Nanyang Lake, Dushan Lake, Zhaoyang Lake and Weishan Lake. Because flood of the lake rising up quickly and falling off difficultly and the complex water exchange relationship with waterfront areas, it is difficult to develop a flood routing model with both simulation accuracy and computational efficiency. The paper uses fourth-order Runge-Kutta method to improve the model revolutionarily based on the idea that flood of the Nansi Lake were generalized to "three-lake" and "two-river" in the 1960s and "three-lake and two-river" flood routing model with auxiliary curve method. The comparative results show that the improved flood routing model is superior to the traditional auxiliary curve method in calculating precision, flexibility and efficiency. In addition, the paper analyzes model sensitivity from the empirical parameters such as drainage modulus and propagation time and the effect of stage-discharge relation of Hanzhuang sluice on peak value of water level of these lakes, and then puts forward some suggestions for the treatment of flood control of the Nansi Lake. The improved flood routing model with "three-lake and two-river" of the Nansi Lake can serve as the basic tool for the flood management of the Nansi Lake. The modeling ideas for local conditions can provide important reference value to similar lakes and reservoirs.