Abstract: Using traditional statistical verification methods and refined verification methods, the precipitation in Sichuan during the 2022 flood season was evaluated based on the CMA-MESO, CMA-GFS, and ECMWF, the performance advantages and bias characteristics of the three models were compared. Furthermore, based on differences in topography and climate, the province is divided into four sub-regions for separate assessment: the Sichuan Basin, the Border Slope region, the Panxi region, and the Western Sichuan Plateau. The results show that: (1) For the province-wide precipitation, the CMA-MESO model shows the best performance for clear/rainy day forecasts and heavy rainstorm forecasts. The CMA-GFS model performs the worst for rainstorm and heavy rainstorm forecasts, with the highest omission rates. The ECMWF model excels in forecasting moderate rain, heavy rain, and rainstorms. (2) For the sub-regional precipitation, the CMA-MESO model delivers the best forecasts for all precipitation grades in the Panxi region. The CMA-GFS model performs best for clear/rainy day forecasts in the Border Slope region. The ECMWF model demonstrates superior performance for moderate rain forecasts in all regions but shows essentially no forecasting skill for heavy rainstorms in the Panxi and Western Sichuan Plateau regions. (3) With the prolongation of forecasting time, both the CMA-MESO and CMA-GFS models exhibit a decline in performance. Forecasts initiated at shorter lead times show reduced false alarms and misses. The models are more stable in forecasting high-frequency events, whereas forecasts for low-probability events are more random. (4) The trend in average precipitation and precipitation intensity in the CMA-MESO model is most similar to that of the actual situation, but the peaks at nighttime are larger. The CMA-GFS model is better in predicting the average precipitation and precipitation intensity in the Western Sichuan Plateau when initialized at 12 UTC. The ECMWF model shows the largest deviation in forecasted precipitation frequency from observations, with the peaks occurring during the daytime while the actual situation is at nighttime.