Abstract: According to early flood season precipitation data(from April to June) from seven weather stations in northern Guangdong province from 1962 to 2006,the relationships between early flood season precipitation and 74 circumfluence indexes were analyzed by grey interconnect degree method.And 16 key general circulation indexes influencing early flood season precipitation in the study area were extracted.Early flood season precipitation models were established by the projection pursuit regression method,back propagation neural network method and stepwise regression method.And early flood season precipitation trends were predicted by these models.The results show that the prediction accuracy from the projection pursuit regression method and back propagation neural network method is much better than that from the stepwise regression method.Among three methods,the projection pursuit regression method is the best one.