Based on the debris flow data collected from 129 villages and towns in the Anning River valley of Liangshan Prefecture,two multi-classification support vector machine models were built to evaluate debris flow hazards of the villages and towns.86 samples from the villages and 65 samples from the towns were randomly selected as training samples and the remainders,as testing samples.Results show that the prediction accuracy of SVM model is improved with the increase of training samples and prediction accuracy of the two SVM models are higher than that of BP neural network models.Therefore,support vector machine method is a new machine learning method with higher precision and better generalization performance than neural network method.It has very broad application prospects and promotion and application values in the practice of debris flow hazard assessment.