Face recognition system is one of many biometric authentication systems which are used for authentication purpose. It provides an effective way for authentication with varied security features. Management of high dimension data is problem which arrives frequently in face recognition. In this paper, a new method is proposed for biometric authentication based face recognition which is general and efficient approach using Radial Basis Function (RBF) kernel to manage small training sets of high dimension. In order to avoid over fitting and reduce the computational complexity, face features are first extracted by the Singular Value Decomposition (SVD) method.