Abstract: The traditional point-to-point binary classification test method can objectively reflect the overall performance of the model forecast, but the method has a double penalty phenomenon. Based on traditional methods, combined with FSS and MODE, the two regional heavy rains affecting Sichuan in July 2021 were evaluated, and the forecast performance of the four models of BCSH, ECMWF, SW3KM and SW9KM were compared and analyzed. The following conclusions are drawn: (1) The TS scores of BCSH and ECMWF models were slightly better than the other two models in terms of light rain to moderate rain. The advantage of the SW3KM model was in the forecast of heavy rain; the BS of the BCSH and SW9KM models had no significant regularity. The ECMWF model had more false alarm from light rain to heavy rain, and the SW3KM model had more false alarm from moderate to heavy rain. (2) The FSS of light rain based on SW3KM model at 72h forecast time, the FSS of heavy rain during 36~72 h and the FSS of rainstorm during 24~66 h were higher than other models; with the improvement of the regional model, the FSS and TS scores increased accordingly. With the extension of the forecast time, the FSS score of the regional model was mainly higher than that of ECMWF model, and the FSS and TS scores of all levels in SW9KM model was lower than that of SW3KM model. (3) The meridional deviation of the precipitation forecasts of the four models was slightly larger than that of the latitude. The centroids of precipitation of BCSH and SW9KM models were in the northwest, and the rainstorm centroid of ECMWF model was in the northwest and the heavy rainstorm was in the southwest, while the centroid of rainstorm of SW3KM model was in the southwest and the heavy rainstorm was in the northwest. The ECMWF model was better than the regional model in controlling the direction and area size of the rainband; the SW3KM model was better than the BCSH model in 72 h forecasting time, and the matching target attribute value of SW3KM model was mainly better than that of SW9KM model; BCSH, ECMWF and SW3KM models had strong forecast of precipitation intensity.