The traditional rule mining algorithm includes attribute reduction and attribute value reduction, which incorporates redundant computation. The complexity of the algorithm will increase dramatically as the sample dataset increases. Therefore, the granular computing(GrC) method is adopted. Firstly, the granular-relation matrices between condition granules and decision granules in different granular spaces are computed. Then the attribute value is reduced according to H ₁ and H ₂ which are hidden in the granular-relation matrices. Furthermore, redundant attributes are removed and the termination condition is set, which can accelerate the mining of decision rules. The theoretical analysis and experimental results show that proposed algorithm can acquire more concise rules, and the rules have better generalizing ability.