Abstract: In the traditional joint sparse model, a dictionary was used for feature representation of the commonor unique components. This leads to the low discrimination of the sparse coding coefficients. In thispaper, mutual information-based multi-channel joint sparse model is proposed for histopathological imageclassification. The training samples are clustered into R, G and B channel dictionaries by using K-means. Byexploring the mutual information between training samples and three dictionaries, the irrelevant atoms aredeleted, meanwhile, a shared dictionary and three unique dictionaries constructed. Simultaneously,multi-channel joint sparse model is designed based on the shared dictionary and three unique dictionaries.Furthermore, in order to represent image feature of different levels, the spatial pyramid matching is used tothe multi-channel joint sparse coding. Finally, the joint sparse coding coefficients are used to train the SVMfor histopathological images classification. The experimental results show that the proposed model haspower feature representation ability and improve greatly the discrimination of coding coefficients. Thus thebetter classification performance and the power robustness can be obtained with compared to the traditionalmodels.