In this paper, fault-tolerant attitude tracking control problem is investigated for multiple spacecraft formation flying system with external disturbance, actuator saturation, and faults. A quaternion-based adaptive fault-tolerant control law is proposed based on input normalized neural network. The desired nonlinear smooth function is approximated by using input normalized neural network with an adaptive learning algorithm, and no prior knowledge about spacecraft dynamics is required. Meanwhile, in order to guarantee that the output of input normalized neural network used in the controller is bounded by the corresponding bound of the approximated unknown function, a modified adaptive law is designed to revise the sliding mode manifold. Moreover, the stability of system can be guaranteed by Lyapunov theory. Finally, the validity of the proposed control algorithm is verified through numerical simulations.