Abstract: In terms of the same predictors of previous monthly mean 500 hPa height and sea surface temperature,a qualitative and quantitative models of long-term forecasting of rainfall over Changjiang-Huaihe Valley are constructed aid of the variable topological structure of neural network.A stepwise regression model is established using same predictors.The contrast analysis shows the comprehensive forecast method of the qualitative and quantitative is superior in improving predictive accuracy and stability.This shows great promise in the research of precipitation prediction on long-range basis due mainly to the reasonable analysis for prediction method and prodictors.