To predict summer precipitation in Chongqing in Southwest China, a downscaling method targeted at the interannual increment of predictand instead of the interannual anomaly of predictand is developed with the Beijing Climate Center Climate System Model (BCC_CSM). Predictions of precipitation, geopotential height, winds, and sea surface temperature by the BCC_CSM and the precipitation observations from 34 weather stations in Chongqing in Southwest China during 1991–2018 are used to establish and validate the method. Specifically, for each of the 34 stations, correlations between the interannual increment of precipitation at the station and the above predicted variable fields in the globe are examined, and the key regions with the highest correlation coefficients are then selected. The predicted variables over these regions are treated as the optimal predictors and are further used to establish three kinds of regression functions for predicting the interannual increment of precipitation. Finally, summer precipitation is predicted by adding the forecasted interannual increment in the target summer onto the observation in the previous summer. Results show that the original precipitation predicted by the BCC_CSM is obviously underestimated in Chongqing. The downscaling predictions, especially the one based on the multivariate stepwise regression approach, achieve reasonable prediction accuracy across years and sites. For the forecasts starting at March 1st, April 1st, May 1st, and June 1st, the skill scores for summer precipitation prediction increase from 80.7, 41.9, 82.8, and 43.5 to 82.5, 66.7, 86.2, and 86.6 in 2017, and from 89.8, 82.8, 55.3, and 85.8 to 91.4, 83.7, 78.1, and 93.2 in 2018, respectively. In addition, the downscaling method could better predict the abnormal-rainfall areas in Chongqing.