For solving the problem of distinguishing different types of diseases when the features extracted from different disease leaves are similar, an automatic recognition algorithm for disease rice leaves using multi-task joint sparse representation was proposed. The proposed algorithm considered the process of recognizing disease by the feature extracted form disease leaf as a task, and required that the feature are came from one task can be sparsely represented, and these sparse coefficient vectors had similar structure. The proposed algorithm can achieve to improve the accuracy rate by correlations among different tasks thus make the recognizing process completed in one-step. The experimental results demonstrated that the algorithm of multi-task joint sparse representation can mine the correlation among the features adequately, and thus the accuracy rate of recognition was improved.