Abstract: The semi-supervised support vector machine(SVM) classification algorithm based on Gauss mixture model kernel is proposed. The unlabeled samples information is provides by constructing the Gauss mixture model kernel SVM classifier. The SVM algorithm is not only study labeled samples information, at the same time, it also can take into account the cluster assumption throughout the training sample set. The comparative experiments are performed with the traditional SVM, transductive SVM and random walk semi-supervised algorithms. The experimental results show that the proposed method not only can improve performance of SVM classification in few training samples, can also increase the overall robust performance.