Abstract: Based on neural networks, a new robust control methodology is presented for nonlinear uncertain composite systems of which the nonlinearities are assumed unknown. The unknown disturbance and interconnected terms are approximated using neural networks. The nominal system is considered. to develop a nominal controller. Then, the unknown disturbance and interconnected terms are taken into account to design correction control signals and add them to the nominal controller, thus the actual system is guaranteed to be uniformly ultimately bounded. In this way the control signals are smooth without the requirement for knowing the upper bounds on the optimal weight values and modeling error in advance. Numerical simulation studies are used to illustrate and clarify the approach.