Abstract: Mining the association rules is an important aspect of the study of data mining. This paper analyzes some problems existing in those available association rules mining algorithms. Firstly, association rules neglect the effect of all kinds of potential negative examples, which results in imperfection in their expressive ability; secondly, it is probable that a rule will have no practical significance or even be misleading even if it possesses high confidence and support. The paper revises the formal definition of association rules, makes use of the deviation based interest value presented in association rules, and defines it in form. In addition to analyzing the practical significance of the interest value, the combination of the interest value and concept hierarchy are also discussed.