Knowledge mining and discovery is a new technology for decision support system beginning in the middle of 1990s. This is a combination of artificial intelligence, machine learning and database technology. To mine knowledge from large databases, the efficiency of leaning algorithms has to be considered. Applying rough set theory to data mining, the attributes relevant to the mining task can be drawn from databases, and simultaneously simple and concise knowledge can be found effectively. The theoretical basis of this application and the realization of the algorithm are presented and discussed.