Based on the causality of information flow, this study examined predictors for the dominant modes of autumn precipitation in Southwest China (SWC), then a statistical model of autumn precipitation in SWC was established. Finally, the prediction skills of the empirical model were evaluated. The first two dominant modes of autumn precipitation during 1979-2020 in SWC are basin mode and saddle mode obtained by empirical orthogonal function (EOF), which are closely related to the developing of eastern El Ni?o and central El Ni?o. The predictors of PCs of the first two dominant modes are chosen to begin with the causality of information flow. The multiple linear stepwise regression with leave-one-out method was applied to further select the predictors and then establish a statistical model. During the training period from 1980 to 2015, the correlation coefficients between predicted PC1 and PC2 and actual PCs are 0.89 and 0.83, and the sign coincidence rates are 90% and 83%, respectively. In the forecast years from 2016 to 2020, the predicted PC1 and PC2 are in phase with the actual PCs in four years, with the sign coincidence rate being 80%. During the 36-year training period from 1980 to 2015, the averaged anomalous pattern correlation coefficients (ACC) between the reconstructed precipitation with the predicted PCs and the observed precipitation anomalies is 0.48. ACC is greater than 0.5 in more than 1/2 years. The regional averaged temporal correlation coefficient (TCC) is 0.48. We also conducted similar-year forecast with predicted PC1 and PC2 to make up for the defect of weak precipitation of the reconstructed field.