Adaptive neural robust controller was designed by using adaptive backstepping method for a class of unknown chaotic system which could be turned to "standard block control type". It was proved by constructing Lyapunov function step by step that all signals of the system are bounded and exponentially converge to the neighborhood of the origin globally. Finally, simulation study is given to demonstrate that the proposed method is effective and the known information of system was made use of as maximally as possible by introducing the PID control.