Abstract: Feature selection is one of the key steps in text categorization, the selected feature subset directly influences results of text categorization. The feature distinguish ability based on word frequency and document frequency is presented. Meta-information is introduced into rough sets and an attribute reduction algorithm based on meta-information is provided. A comprehensive feature selection method is proposed. The comprehensive method firstly uses the feature distinguish ability to select feature and filter out some terms to reduce the sparsity of feature spaces, and then employs the provided attribute reduction algorithm to eliminate redundancy, so that the more representative feature subset is acquired. The experimental results show that the comprehensive method in a certain extent has advantages.