Abstract: A design process is presented for a class of nonlinear composite systems to which the dynamic neural network is used to approximate. Nonlinear composite systems are identified by means of dynamic neural network, i.e., use dynamic neural network to approximate to the unknown and interconnected terms of a system. Then design a controller to enable state of the actual system to track the locus of reference model. Based on Lyapunov's stability theory, the tracking error and other signals are proved uniform and bounded eventually. The practicability of such a design process is verified by a simulative example of nonlinear system. A conclusion is drawn that the design process can solve most complicated problems of interconnected terms in a composite system. A law of adaptive control is suggested on neural network basis.