| Non-point Source (NPS) pollution of urban runoff is one of the major causes of water degradation, setting up initial rainwater storage tank is one of the most economical and effective approaches to alleviate the runoff pollution. At present, the initial rainwater storage tank is usually designed by means of the empirical method, which does not consider the factors as the characteristics of pollutant accumulation in catchments, the sewage treatment plants capacity, and the available land etc., resulting in limited pollutant interception effect. In this paper, the characteristics of accumulated runoff pollutants was simulated by means of SWMM model, and a high-precision nonlinear mathematical model between the pollution interception rates and the corresponding interception runoff volume for the initial rainwater storage tanks was obtained by means of the ANN (Artificial Neural Networks). Combined with the model, the optimal pollution interception rates of the storage tanks was obtained by means of the PSO (Particle Swarm Optimization) algorithm with the maximum pollutant interception amount of the storage tanks under the constraints of sewage treatment plant capacity and the available land. And finally, the volume of the storage tanks was calculated according to the relationship between the pollution interception rate and runoff accumulation. This method was also compared with the empirical design method widely used in China. The results showed that the method proposed in this paper could maximize the interception effect of the storage tank under the constraints, and the storage tank designed by this method was superior to that of the empirical design method in terms of land utilization rate and pollution interception effect. This method is scientific, efficient and widely applicable, which can provide technical support for the design of the initial rainwater storage tank.