The decision information system of off-line handwritten Chinese character recognition based on variable precision rough set is constructed in this paper. The granular variable in granular theorem is introduced, the granular entropy of feature attributes knowledge and decision attributes knowledge and relative granular entropy is defined. The attribute reduction algorithm with several features is given out. Moreover, the recognition structure which has three-tier classifier based on rough set iterated several times is designed. And it resolved the problem of high rate of rejection and coincident code with a single classifier and recognition only once effectively. The experiment results show that the method is feasible and effective.