For a robot system with model uncertainties and unknown nonlinear perturbations, an adaptive fuzzy sliding mode decentralized control method was put forward by considering the upper bound of the uncertainties and unknown disturbances to be ordinary high order polynomials about the system states, and combined with the approximation ability of fuzzy systems. This method can not only make every joint’s controller of robot systems with mutual coupling between the joints completely determined only by the information of itself, but also eliminate the related assumptions which the extant literatures in the design of robot decentralized controller requires modeling uncertainties and the upper bound of the unknown disturbances to be a constant or firstorder polynomial about system states. Lyapunov stability theory analysis shows the closedloop system is semiglobally uniformly ultimately bounded with tracking error converges to zero. Simulation results of two degrees robot systems prove the effectiveness of this method.