We propose an efficient and accurate classification method based on Sparse Representation based Classification (SRC) for face recognition. In this approach, instead of using all or a subset, we use cluster centers of training samples to build SRC models. Considering the variability and redundancy of training samples, each class will be represented by a different number of representatives. In the next step, different feature vectors are extracted from this abstract training set and different modalities are formed which are then used in a multimodal sparse representation framework to classify unknown test samples. Face recognition experiments on two different face datasets confirm the proposed multimodal method has higher recognition rates in comparison to single-modality methods. The proposed method is also compared to other multi-modality classifiers and results confirm that higher recognition rates can be achieved with this method.