Abstract: To deal with the problem that continuous value can't be applied to the rough set theory(directly),a new attribute discretization approach applying Parzen windows and genetic algorithm(GA) is proposed.Parzen windows which can describe the data distribution status provides stability index of the results for continuous value discretization.Many cutpoints are chosen to classify the continuous values at the beginning,then the consistency requirement of rough set theory and the stability index of Parzen windows are used to define the fitness function of GA.The simulation results show that just a few cutpoints can keep the consistency of knowledge base classification and the stability index is good as well.