Dynamic equations of robot manipulators are highly nonlinear with time-varying and unknown parameters. Using the model reference adaptive control (MRAC) technique, a control scheme based on hyperstability theory is developed for robot manipulators. A new adaptive algorithm is proposed for compensating the nonlinear term in the dynamic equations and for decoupling the dynamic interaction among the joints. The main feature of the approach is that the unknown parameters are not estimated separately, but the total influences due to the modeling errors and the disturbances can be directly compensated. Simulations show good results even for large variations of parameters. A comparison of this approach with the feedback linearization method (FLM) is also presented.