Abstract: Association rules is an effective method for describing the dependency relations in data, and it is one of the improtant aspects of knowledge discovery. Taditional association rule mining methods lack of focus on the results, and the procedure is slow. Those algorithms express the regularities with low level primitive data, and the mining association reules are difficult to understand. Furthermore, the desirable knowledge must be filtered out from huge results in a post\|processing step. A method for integrating concepts into the mining procedures to improve the interestingness of results and speeding up mining procedures is proposed, and the method for deriving concepts interactively in large database is proposed, a concept\|guided association rules mining algorithm CGARM is given in the paper here. CGARM extends taxonomies\|based mining methods. Experiments show the execution speed of CGARM is about twice as faster as the traditional mining algorithm Cumulate, and the interestingness of results are also improved.