Abstract: In this paper, we discuss the learning control for nonlinear discrete system with delay. It is not necessary to use the desired input corresponding to the desired output and the desired state corresponding to the system in our learning algorithm and in the process of learning control. Therefore, our method needs less dynamics information of the controlled object than previous ones and only the qualitative Lipschitz condition is required. Not only is the learning algorithm convergent, but also it can track the desired target in common sense (not tracking the desired target only in a neighborhood as the existing ones). Our algorithim includes some existing ones as its special cares.