Abstract: Based on three climate model products from NCC_CSM11, NCEP_CFS2 and ECMWF_SYSTEM4, atmospheric circulation reanalysis data from NCEP and monthly precipitation data of 359 stations in Yunnan, Guizhou and Sichuan provinces, seeking "key area with high prediction skills" of models and implementing stepwise regression scheme, the performances of monthly precipitation prediction of model interpretation results using interannual increment data and source data are discussed compared to model direct output results. The conclusions show that the PC score of model interpretation results using source data is close to that of model direct outputs, while the PS and ACC scores are slightly improved. The performances of model interpretation results using interannual increment data are superior to those of using source data. Furthermore, the integrated prediction results are obtained based on superensemble mean integration method with three model interpretation results using interannual increment data each month. In most cases, the prediction scores of the integrated result are higher than the maximum score of the model interpretation results involved in ensemble process. The improvement of ACC of integrated result is the most obvious, followed by PC, while the improvement of PS is not clear.