In order to increase the forecasting accuracy, techniques of sh ortseries modeling and residual correction are improved and the prediction mod els of original series, shortseries and corresponding residual correction are developed on the basis of prediction technique of averagedgeneration functions. The used data are the precipitation data of Baishe City from June to August in 195120 06. The results show that the forecasting accuracy and precise of these improve d models have raise to some extent, especially the Ts of MGFAFM model is 25% hig her than MGFFM. The operational use of a consensus forecast with four models to the monthly rainfall amount is a good choice.