Neural network control for robot manipulators is aimed to compensate for uncertainties in the robot dynamics. The location of a compensating point differentiates the control scheme into two categories, the feedback error learning (FEL) scheme and the reference compensation technique (RCT). The RCT scheme is relatively less used although it has several structural advantages. In this paper, the global stability of the RCT scheme is analyzed on the basis of Lyapunov function. The analysis turns out that the stability depends upon the magnitude of the controller gains. Simulation studies of controlling the position of a two-link robot manipulator are conducted.