In order to improve the capacity of nonlinear fitting and accuracy of flood forecasting, the support vector machine model was studied. For the actual situation of the Linjiang Station on the Yalujiang River, the flood forecasting model for the Linjiang Station with ν-SVR forecasting model was established. The precipitation and discharge process data in the flood years from 1998 to 2014 in the watershed were used to calibrate and validatethe the model. The flood forecasting model for the Linjiang Station with ν -SVR was compared with the linear dynamic system model, BP neural network model and ε -SVR model. The results show that the flood forecasting model with ν -SVR has a high precision than the linear dynamic system model and BP neural network model. The flood forecasting model with ν -SVR has better nonlinear fitting ability and generalization ability, can control the number of support vectors and reduce the complexity of the model, while maintaining a good prediction accuracy.