Sparse representation in compressed sensing is a recently developed hot research area in signal processing and artificial intelligence due to its success in various applications. In this paper, a new approach for facial expression recognition (FER) based on fusion of sparse representation is proposed. The new algorithm first solves two sparse representations both on raw gray facial expression images and local binary patterns (LBP) of these images. Then two expression recognition results are obtained on both sparse representations. Finally, the final expression recognition is performed by fusion on the two results. The experiment results on Japanese Female Facial Expression database JAFFE show that the proposed fusion algorithm is much better than the traditional methods such as PCA and LDA algorithms.