Support Vector Machine(SVM) has good performance for classification,but the performance is restricted by the kernel function and its parameters.This paper discusses the problem,and uses cross validation,grid searching for optimizing the kernel function parameters.