A new algorithm is presented for face recognition based on DCT and a multi-class support vector machine (SVM) model. The extracted features from human face images by DCT have major information that can be recognized. As a classifier, the SVM has its particular advantage in tackling small sample size, high dimension and etc., and is of high generalization and without need of priori knowledge. The results on ORL face database show that the DCT feature extraction method is effective and the SVM is superior to the nearest neighbor classifier in classification performance with the efficiency of the whole system improved.