The prediction models of droughts and floods in flood season over Zhejiang Province are built based on the Support Vector Machine (SVM) method. By use of rainfall data from 38 observation stations, the indexes of drought and flood intensities are devised, which can represent the overall state of droughts and floods in Zhejiang Province. Taking the indexes as predictands and antecedent atmospheric circulation and SST, which exhibit high correlation with the predictands, as predictors, the prediction models of droughts and floods for Zhejiang based on the and regression method are built, respectively. The comparison between the prediction results by two methods shows that the SVM prediction model could make use of the plentiful predictor’s information and nonlinear projection capability effectively, and shows better performance,which was confirmed by both training and testing samples.