Due to the good learning and generalization performance, the SVM （support vector machine） has been widely used in practice. But, how to make the SVM more effectively perform incremental learning is a problem that needs to be solved in the present application of the SVM. The distribution characteristics of Support vectors are studied and a novel improved incremental SVM learning algorithm-distance ratio algorithm is proposed. According to the removing rules of the proposed method, an appropriate parameter is set and samples that have less effect on later training are abandoned. According to the definition in distance ratio algorithm, the ratio between the center distance of each sample and the distance of each to the optimum classification surface is calculated. In this way, the training data sets can be effectively reduced. Experiment on standard data sets shows that by using this method the classification accuracy can be guaranteed and the training speed can be effectively improved.