2D image-based face recognition techniques are sensitive to changes in illumination, pose, facial expressions and make up. We present multi-modal hybrid face recognition which exploits the texture and 3D shape of the face for accurate person identification. The pose of a textured 3D face is automatically corrected using an iterative algorithm based on a single automatically detected point and the Hotelling transform. Face recognition is performed in three stages. In the first two stages, unlikely candidates are rejected leaving fewer for the next stage. The Spherical Face Representation (SFR) is used in conjunction with the SIFT to form the first rejection classifier. The second classifier detects keypoints on the 3D face and extracts invariant local features at the keypoints. The local features are projected to the PCA subspace and matched to find similarity between two faces using a graph-based approach. Similar faces have more matches and result in a similar graph (see figure below). Most of the time, faces are recognized in the firs two stages and only challenging ones need to be verified by the third stage. The third stage makes the final decision using region-based matching of the expression insensitive regions of the face.