Abstract: Based on several levels data of monthly NCEP circulation data,the prominent correlative regions have been chosen.Through filtering and association with the optimal subset regression method,35 stations of Anhui monthly precipitation regression models have been formed and 2005-2009 monthly precipitation in Anhui have been experimentally predicted with the incoming data of dynamic extended range forecast(DERF) model.The results show that the downscaling model contains multiple data of low-high levels,various factors have been assembled to show some allocation situations affecting precipitation,which are more comprehensive than the model only using 500 hPa data.Seeing from experimental predication results,mean anomaly concordance rate is 63%,the PS score is 75.The method is preponderant in waterlogged month.