Abstract: Some key problems in face recognition are studied in this paper. A new method of face detection,a new SVD-based method of feature extraction and a new method of face recognition based on SVM are proposed. The improved method of feature extraction based on SVD can extract face features better than the traditional PCA. Only one SVM is built to solve face recognition,which is a typical problem of multi-classification,thus overcoming some flaws of several traditional multi-classification methods of SVM. The experiments are done on the FERET,BioID and the face databases with manual as well as automatic face detection means. The results show that this new detection method presents high correct rate. The new feature extraction method based on SVD weakens the influence of illumination and expression on recognition in order that the higher rate of correct recognition is obtained. The new face recognition method based on SVM has better capability of generalization and higher rate of correct recognition than other SVM methods. These new methods provide academic and experimental warrant for building an automatic face detection and recognition system based on SVM.