Abstract: Text classification is an important field in data mining and machine learning. In recent years, the use of association rules for text categorization has attracted great interest and a variety of useful methods have been developed. These works focus on how to generate classification rules and then pick rules to build a high accuracy classifier. By analyzing association-rule based text classification, an observation may be obtained that decreasing error classification for negative samples may improve classification accuracy while keeping categorizing positive samples unchanged. Inspired by negative selection algorithm, the authors propose a classification rule revising strategy to implement the above observation. First, a new rule, called negative rule, is generated by mining frequent item sets on negative samples that are error categorized by a classification rule. Then the classification rule is combined with its negative rules to generate an enhanced association rule. The enhanced association rules can dramatically decrease error categorization for negative samples, and therefore classification accuracy is improved. Experiments are conducted on a real Web pages dataset. Compared with text classification algorithms (CMAR, S-EM and NB), the rule revising strategy may further improve classification accuracy. The utility and feasibility of the revising rule strategy are also demonstrated by formalization proof.