Abstract: A junk messages filtering method, based on an algorithm of support vector machine (SVM), is proposed. By means of performance evaluation index universally applied in such fields as text classification and information retrieval, an evaluation system for filtering junk messages is established. On the basis of data obtained in stimulation and by means of evaluation results of the evaluation system for filtering junk messages on experimental data, the proper kernel function and its parameters are selected to construct a classifier of support vector machine; The classifier of support vector machine and the traditional Bayesian classifier are tested and evaluated through stimulation and evaluation system. The comparison results show that the classifier of support vector machine improve the accuracy of filtering messages, and verify the effectiveness of the algorithm of support vector machine in junk messages filtering.