A global optimizing training algorithm based on Tabu search for SVM is proposed,which improves the SVM classifier' s accuracy when the samples with low dimension feature vectors.Simulation indicates that the proposed SVM classifier' s ability proves to be good.