Abstract: This paper presents a fuzzy support vector machine forecasting method based on gray correlation analysis for forecasting streamflow featured with nonlinearity and randomness. This method takes advantage of traditional SVM in its arbitrary approximation ability and nonlinear mapping, and adopts a fuzzy membership function to consider the impacts of changes in climate and watershed surface conditions on streamflow forecasting results. Predictor selection is difficult in long-term streamflow forecasting and the correlation coefficient method can only measure linear correlation between factors. Hence, we adopt gray correlation analysis to quantify the degree of association and pick out predictors that have significant impact on the streamflow. This model was applied to forecasting of monthly stream flow at Shigu, a control station of the upper Jinsha river. Comparison with the GRNN model and A-FSVM model shows that the method is effective and improves the accuracy of long-term streamflow forecasting.