Abstract: Using monthly dynamic extended range forecast (DERF) products, NCEP/NCAR reanalysis data and 160 station data in China, we have extracted optimal predictors which significantly influence precipitation. The predictors are selected from high skill regions of DERF and high correlation between observing precipitation and other variables. Downscaling monthly anomaly precipitation are predicted using the optimal subset regression from selected predictors. Two schemes are designed in the paper. One uses the detrending data to establish the prediction equations (scheme 1), the other uses the original data to establish the prediction equations (scheme 2). The results show that the two schemes can improve the skills of monthly precipitation forecast whereas scheme 1 has higher skills than scheme 2. The anomaly correlation coefficient between scheme 1 and observation can reach 0.35. Downscaling tests indicate that initial fields could have great influence on model results and downscaling precipitation prediction. Using scheme 1 and DERF results from different leadtime including 0 day, 5 days and 10 days, it shows that when lead time is 5 days and 0 day, the downscaling precipitation prediction could be used as a good reference in monthly climate prediction. In the downscaling model, selected predictors can pass statistic tests and have clear physical meanings. The downscaling methods can be applied in operational climate prediction.