Multi-granularity is one of the important research directions in the field of granular computing. It represents the research of problem solving at two or more different granules and already has become a new computing method to solve complex problems. Rough set theory is a kind of computing tool to solve uncertain problems effectively. As one of the core contents of rough set theory, attribute reduction has been widely studied in the fields of data mining, machine learning, and granular computing. It is a meaningful problem to apply the idea of multi-granularity to attribute reduction. The granulation idea in granular computing theory is used to granulate attributes to form multiple granules. Then we evaluate the significance of attribute granules and the significance of attributes in attribute granules based on discernibility matrix. Finally, a multi-granularity attribute reduction algorithm based on these two significance measures is designed. By selecting attributes from different granules, the reduction results obtained by this algorithm are more representative and different. In order to verify the effectiveness of our proposed method, experiments on six data sets show that the effectiveness of the proposed algorithm.