A pose-invariant face recognition based on features selection and statistical classification is proposed. The proposed method includes training and testing phases. SURF features are utilized to calculate the similarity between two images from different poses of the same face, in training phase. Then, Gaussian Mixture Models (GMM) are trained using the robust SURF features from different poses. For the decision of face recognition, feature vectors corresponding to the test images are fed to all trained models. Experimental results demonstrate the proposed method outperforms the existing techniques.