Abstract: Based on the daily precipitation forecast data for the next 45 days with NCEP CFSv2 model and station observation data, the prediction skills for monthly spatial patterns and interannual variations of precipitation over southwest China were evaluated. The results show that except the wet bias in the northeastern Sichuan and northeastern Chongqing in rainy season (from May to September), there are almost dry bias in other regions and months. The months with significant bias are mainly concentrated in summer. The regional average bias shows an upward-downward-upward trend from January to December, with the highest in April and the lowest in October. As the leading time increases, the prediction skills of spatial patterns decrease, but the prediction skills of magnitudei increase. In different leading times, the “Ω” trend prediction of the model in moderate rain days, heavy rain days and rainstorm days is basically correctly, but the light rain days show opposite characteristics with observations. The prediction of light rain and moderate rain days is higher than that of observation, but the forecast of heavy rain and rainstorm days is lower than that of observation. In interannual variations, the model is predicted to be positive correlation in most areas, while negative skills are mainly concentrated in Sichuan Basin to the western Sichuan Plateau and the border areas of Chongqing, Sichuan, and Guizhou provinces. The high correlation coefficient region reflects the process of gradually pushing northward and westward from spring, summer, autumn and winter, and then retreating eastward and southward. However, the magnitude of root mean square error (RMSE) is distinct in different months and regions. For the interannual variation of regional average precipitation, there is basically positive correlation coefficient, and the correlation coefficient is higher as the forecast time is approaching, but the RMSE improvement is not obvious.