The number of rules in belief rule base(BRB) may induce the problem of combinatorial explosion. However, most previous works on rule reduction are based on feature extraction, whose effectiveness depends on the expert knowledge. Therefore, an objective method based on rough set theory is proposed, which does not depends on any knowledge in addition to belief rule base. The method of rule reduction analyzes each belief rule according to equivalence class division thought,
and then eliminates the redundancy of referential values. Finally, a numerical case study to evaluate armored system is analyzed and compared with the typical subjective method in the number of reduced rule and the accuracy of decisionmaking. The results show that the proposed method is feasible and effective, and superior to the existing subjective method of rule reduction.