This paper studies a new learning observer-based fault-tolerant tracking control strategy for a hypersonic vehicle attitude system. It takes into account the external disturbance, parameter uncertainty and actuator faults. Firstly, a new learning observer combined with a barrier function is proposed to estimate the attitude angular velocities and reconstruct lumped disturbance accurately and quickly without obtaining the prior knowledge of lumped disturbance. Then, by introducing the estimation value into the controller design process, a back-stepping fault-tolerant tracking controller is developed to solve the attitude tracking problem of hypersonic vehicle. And the stability of a faulty attitude system for hypersonic vehicles based on such control method is analyzed by using the Lyapunov framework. Finally, the feasibility and superiority of the fault-tolerant control scheme is proved by comparing with different existing observer and control strategy through numerical simulations.