Fitting precision and generalization ability of support vector machines (SVM) depends on the nuclear parameters of kernel function and punish parameters largely. Researches on Chinese classification experiments results indicate that FOA-SVM can achieve high precision of modeling and the higher classification accuracy.