Nowadays，there are many models for river runoff prediction，but each model has its advantages and disadvantages， and there is difference in prediction accuracy． Taking Nanshui river in Shaoguan as the research object and using the runoff data of Nanshui river from 1978 to 2015，the grey system theory，BP neural network based on genetic algorithm ( GA － BP) and support vector machine( SVM) are used to predict the annual runoff of Nanshui river espectively． The results show that the prediction accuracy of BP neural network based on genetic algorithm is 84． 30% ，but its fitting accuracy is not high; the prediction accuracy and the fitting accuracy of grey system are 86． 70% and 84． 24% respectively; whether the fitting accuracy is 95． 67% or the prediction accuracy is 99% ，the prediction of annual runoff of Nanshui river by SVM is better than that by grey prediction and GA － BP neural network． Therefore，SVM can be well applied to the prediction of annual runoff in Nanshui Ｒiver．