Abstract: An adaptive RBF neural networks based dynamic surface sliding mode control method was proposed for heliostat in solar power tower plant. The "explosion of complexity" in the traditional backstepping method was avoided. The RBF NNs was used to approximate the unknown parameters and uncertainties in the system. By introducing the sliding mode method into the controller design procedure, the robustness and the response of the system was improved. By using the error transformation function, a new error variable was introduced to make the output of the system satisfy the prespecified performance. The hardware-in-the-loop experimental results show that the tracking errors are always satisfy the prespecified performance index. Finally, satisfactory control performance is achieved in the heliostat control system.