Abstract: The paper proposes a multi-classified method of Support Vector Machine（SVM） based on Partial Least Squares（PLS）.The partial least squares is used to complete the pretreatment of the sample，it can eliminate the correlation of the samples’ attribute，and it can also maximize the correlation between the general components and the information of kinds.Through the result of the experiment，we can see that the method can not only reduce the amount of the support vectors in the process of the classification with support vector machine，but also heighten the accuracy ratio of the recognition when the amount of the samples’ attribute is larger.