Paddy runoff pollution is one of the major contributors to limiting the improvement of water quality in Taihu Lake Basin. Grass buffer strips (GBSs) are an effective measure to control paddy runoff pollution. However, most studies only consider a single inflow condition, and few studies have considered the effect of high-frequency rainfall. In this study, a field runoff simulation experiment was constructed to simulate the effect of GBSs on runoff nitrogen and phosphorus removal at different inflow volumes, inflow velocities, inflow concentrations, and rainfall frequencies. Results demonstrated that the larger the inflow volume, the faster the inflow velocity, and the lower the inflow concentration, the higher the runoff pollutant interception rate that occurred in GBSs, and the interception rate improved significantly with increasing GBS widths. The peak change point of removal rate occurred at a width of 15 m for NO3−-N and TP and at a 25 m width for TN and NH4+-N. The cumulative removal rate increased slowly after the change point. Although the peak cumulative removal rate appeared at a GBS width of 35~45 m. Considering the pollutants intercepted by GBSs and the emerging demand for land in this basin, 25 m was recommended as the optimum width to remove runoff pollutants.