The algorithm about incremental generating decision function is built by modifying the definition of discernibility matrixes. So it is programmed effectively in the computer. The discernibility matrixes is not generated because the great quantity of data is fetched one after another. The lack of memory is solved by handing data in large database. The algorithm is adapted to the dynamic data and is able to deal with dynamic learning. The time of Data Ming decreases because of information reuse. At the same time, the question of learning about multi-classes is solved in essence by using the incremental generating decision function.