By representing the input testing image as a sparse linear combination of the training samples via l1-norm minimization, sparse representation based classification (SRC) has shown promising results for face recognition (FR). Particularly, by introducing an identity occlusion dictionary to code the occluded portions of face images, SRC could lead to robust FR results against face occlusion. However, the l1-norm minimization and the high number of atoms in the identity occlusion dictionary make the SRC scheme computationally very expensive. In this paper, a Gabor feature based robust representation and classification (GRRC) scheme is proposed for robust FR. The use of Gabor features not only increases the discrimination power of face representation, but also allows us to compute a compact Gabor occlusion dictionary which has much less atoms than the identity occlusion dictionary. Furthermore, we show that with Gabor feature transformation, l2-norm could take the role of l1-norm to regularize the coding coefficients, which reduces significantly the computational cost in coding occluded face images. Our extensive experiments on benchmark face databases, which have variations of lighting, expression, pose and occlusion, demonstrated the high effectiveness and efficiency of the proposed GRRC method. ; Department of Computing