This paper is for refining database in the data mining process. Based on the analysis of the features and disadvantages of this decision tree algorithm and the substantive characteristics of data mining, we propose the concept of the core samples set and prove its invariance. On this basis, we build an attribute reduction method based on decision tree algorithm and then give a specific implementation steps, further, combined with a specific instance analyze the characteristics and efficiency of the method. Results show that the attribute reduction method based on the decision tree has good maneuverability and explicableness. This method can simply realize the attribute reduction of information system and its basic ideas completely adapt to the attribute reduction problems of the uncertain environment.