In this paper, a design technique of cerebellar model articulation controller (CMAC)-based fault-tolerant control (FTC) system is investigated to deal with the nonlinear faults of the biped robot. The proposed CMAC-based FTC (CFTC) scheme contains two main components: (1) the online fault estimation module based on a CMAC is used to provide the approximation information for any non-nominal behavior due to the faults of the biped robot; and (2) the controller module consists of a computed torque controller and a robust fault-tolerant controller. In the controller module, the computed torque controller reveals a basic stabilizing controller to stabilize the system and the robust fault-tolerant controller is utilized to compensate for the effects of the system failure so as to achieve the fault accommodation. The adaptive laws of CMAC are rigorously established based on the Lyapunov function so that the stability of the CFTC system can be guaranteed. Finally, simulation results show that the CFTC can apparently recover the control performance for the biped robot in the existence of the nonlinear faults.