What the query sub system of a information system provide is large amount of original data, but data is not information, the regularity and rule hiding in data is the information which is useful for decision making. So the data mining technology, which can discover these regularities and rules from data automatically, is an important method of realizing intelligent decision support system. Mining characteristic rules in relational databases is an important content of data mining. Many of current methods of data raining can discover the characteristic rules related to a specific concept，but they are not effective for flexible query．This problem can be solved by using attribute—oriented induce algorithm ，but this algorithm can not guarantee the optimum precision of rule．In this paper，by integrating the theory of rough sets，a method of measuring the rough—degree of data set is presented，and a new attribute-oriented induce algorithm of mining characteristic rule is developed，which is not 1imited by the condition of query．The characteriatic rule mined by this algorithm can describe the characteristics of a concept effectively．On the basis of this new algorithm．a SQL—like language of querying characteristic rule is deslgned，which can make the interaction between user and system more convenient．