Abstract: Based on the existed research about the attribute values have hierarhcal structure in rough-granular computing, this paper proposes a new algorithm for extracting multi-hierarchical decision rules.The aim of the algorithm changes the strategy of reduction, and improves the effectiveness of computing. First, this algorithm use concept hierarchy to organize the attribute values and solve the different depth problem between left nodes and their corresponding righ nodes in hierarchical structure. Next, using the property that the positive region of the decision table in the high level is included its corresponding lower levels, the algorithm constructs the certainty factor matrix. Based on this matrix, it uses distribution reducts to obtain multi-hierarchical decision rules from top to down. Finally, the feasibility of the algorithm is validated by an example.