Abstract: In this paper a random center distance sorting-based support vector pre-extracted method is proposed.For linear separable case,the method firstly randomly selected from each category a certain number of samples and calculated the mean of them as a random center of the category.Then the method calculated for each sample the distance between the sample and the random center of the other class.Finally select a certain number of original samples with small distances to compose boundary samples.For non-linear separable case,the method maps the original problem into feature space with kernel functions,and solves the problem like linear separable case.As support vectors often locate in the border area adjacent to the other type of samples,this method can pre-extract support vectors more exactly.Experiments on UCI standard data sets and ORL data set show that the proposed algorithm can pre-extract support vectors faster and more exactly than previous support vectors pre-extraction methods.