The neural network adaptive reconfigurable control for uncertain nonlinear system is studied in this dissertation, and a set of control methods are proposed for uncertain nonlinear systems with different characteristics.First, the concept of association degree is proposed based on fuzzy logic, and further a self-organizing fuzzy CMAC (SOFCMAC) neural network whose parameters and structure is self adjusted is developed as an improvement of CMAC. It is proved that the approximation to nonlinearity provided by the SOFCMAC can be made arbitrarily accurate.For nonlinear systems with known linear part and unknown nonlinear part, the combination of SOFCMAC neural network with linear control theory results in two effective control approaches, H" robust SOFCMAC neural network adaptive control and SOFCMAC neural network adaptive reconfigurable tracking control, to solve the control problems of such kind of nonlinear systems. SOFCMAC in these schemes is used to cancel unknown nonlinearity to get desired closed loop responses. The proposed schemes are able to effectively extend the applications of linear control methods.For nonlinear systems with partially known nonlinearity, an adaptive SOFCMAC neural network reonfigurable control method based on approximate dynamic inversion is presented. SOFCMAC here is used to cancel the inversion error to make system responses follow the responses of reference model.Further the idea of pseudo-control hedging is applied to the scheme of adaptive SOFCMAC neural network reonfigurable control based on approximate dynamic inversion to solve the control problem of systems with input saturation or digital input. An important point of this scheme is that system structural faults can be self-repaired. And in this self-repairing system, the reconfiguration of control law is not based on the results of fault diagnosis. The results of fault diagnosis are used for alarm and maintenance or play auxiliary role in control systems. The speed requirement of fault diagnosis and identification can be relaxed. A way of fault diagnosis for nonlinear dynamics systems based on support vector machine is presented.The problem to control uncertain nonlinear systems whose states are not available is also addressed here, and a method of direct output feedback SOFCMAC neural network control based on approximate dynamic inversion are presented.The stability of the closed loop nonlinear systems with proposed methods is proved. A series of simulations, such as fighter aircraft, etc., show the effectiveness of the proposed methods.