In this paper, an adaptive neural network multiple models sliding mode controller for robotic manipulators is presented. The proposed approach remedies the previous problems met in practical implementation of classical sliding mode controllers. Adaptive single-input single-output (SISO) RBF neural networks are used to calculate each element of the control gain vector; discontinuous part of control signal, in a classical sliding mode controller. By using the multiple models technique the nominal part of the control signal is constructed according to the most appropriate model at different environments. The key feature of this scheme is that prior knowledge of the system uncertainties is not required to guarantee the stability. Also the chattering phenomenon is completely eliminated. Moreover, a theoretical proof of the stability and convergence of the proposed scheme using Lyapunov method is presented. To demonstrate the effectiveness of the proposed approach, a practical situation in robot control is simulated.