A real-time face recognition method using Gram-Schmidt Orthogonalization for linear discriminant analysis (GSLDA) is presented in this paper. The GSLDA algorithm avoids the large matrices computation such as computing the inverse or diagonalization of matrices, which may be somewhat problematic in terms of computational demands and numerical accuracy. On the other hand, GSLDA also achieves better recognition performance than the classical linear discriminant analysis (LDA) by overcoming the degenerate eigenvalue problem of LDA. Experimental results on real face databases have confirmed the better performance of the proposed method.