A neural networks adaptive back stepping control method was put forward to reduce the influence of all kinds of uncertain factors on robot control system. The control method was based on back stepping and solved mismatched uncertainties of robotic system. Through introduction of neural network adaptive control law and robust control law, and adjusting the controller parameters appropriately, the control could be realized without accurate model of robot. The stability and convergence of the system were also proved and the control requirements of robotic system were met. The effectiveness of the control method was validated by simulation experiments.