Abstract: Attributes reduction is one of important parts researched in rough set theory.This paper weights the frequency of the attributes in the condensing tree，and combines with the heuristic method of the attributes’ frequency weight，takes the core as foundation，deletes the most important attributes in the tree until the tree is null；in order to find optimum reduction of information system，the paper adds the converse eliminate action until cannot delete.A demonstration is given at last in the paper，and it verifies the method validity.

Key words: rough set, discernibility matrix, attribute reduction, condensing tree

摘要： 属性约简是粗糙集理论的重要研究内容之一，对浓缩树结构中属性出现的频率进行加权，以属性频率的权重作为启发，以核为基础，从树中删除属性重要性最大的属性结点，直到树为空；为了找到信息系统的最优约简，在此基础上加了一个逆向消除的过程，直到不能再删为止。最后通过一个实例完整演示了该方法，证实其有效性。

关键词: 粗糙集, 区分矩阵, 属性约简, 浓缩树