Abstract: A wavelet support vector regression-coupling (WSVR) model as an integration of discrete wavelet transform (DWT) and support vector regression model (SVR) was developed to predict monthly runoff. This model uses the Mallat algorithm to decompose a given series of monthly runoff into sub-series of different time scales and reconstructs it, and then inputs the effective sub-series into the SVR model. It evaluates the accuracy in terms of RMS error (RMSE), mean absolute error (MAE), deterministic coefficient (DC), and correlation coef?cient (R). It was applied to forecasting of monthly runoff at the Zhangjiashan hydrologic station on the Jinghe River. The results show errors in the calculations of verification period, RMSE=12.5m3/s, MAE=7.74m3/s, DC=0.87, and R=0.935, a better accuracy in comparison with the optimized SVR model's errors of RMSE=27.9m3/s, MAE=13.43m3/s, DC=0.34, and R= 0.662. Thus, the WSVR coupling model improves monthly runoff forecasting.