Abstract: In this paper, aiming at the redundant rules produced during multi-level association rule mining, an approach for eliminating multi-level redundant rules based on semantic dividing is proposed. In the method, the redundancies are divided combination with semantic ontology into three categories: sub-rules redundancies, synonymous redundancies and reducible redundancies. Deleting the redundant sub-rules, grouping the synonymous rules and treating these rules having the same meaning, reducing rules that express only part of information to higher level, when pruning redundancies. Finally, the approach is evaluated on a real mobile subscription service dataset, and its effective in treatment of the redundant rules is proved through the experiment.