Abstract: Runoff prediction plays a key role in development and management of regional water resources. Previous research focused largely on advanced algorithms, often ignoring the contribution of expanding predictors to runoff prediction. This study has developed an improved support vector machine model (GA-SVR) coupling the genetic algorithm (GA) and a support vector regression (SVR) model through a case study of predicting the runoff in the Jing River. In addition to the conventional forecasting factors (i.e. rainfall and evaporation), predictor variables in this model also cover anomalous atmospheric circulation factors that have a strong influence on the runoff. Results indicate that the GA-SVR model achieves a prediction accuracy and generalization ability significantly better than the neural network model (ANN) when not using these anomalous factors and when putting them into use, the accuracy is further improved. Thus, accuracy in monthly runoff prediction can be improved by coupling a SVR model with GA and further improved by considering anomalous atmospheric circulation factors.