Complexities in controller designs for flexible manipulators, mainly arising from the nonlinear vibrational dynamics, get further compounded when task executions are to be completed in the face of changing payloads. The development of a neural network-based scheme for adaptively implementing a variable structure controller to drive a flexible manipulator arm is described in this paper. The controller provides a satisfactory suppression of the tip vibrations while facilitating a rapid hub rotation for executing motions during which payload variations can occur. An efficient integration of the operational strong features of a trained neural network and a variable structure controller is made in the development of the overall control scheme. On the one hand, the high degree of inherent robustness of the variable structure control serves to reduce the architectural and training complexity of the neural network, while on the other, the faster processing capability of the neural network is exploited for on-line payload identification and adaptive implementation of the variable structure controller. The resulting control strategy is novel in facilitating a trained neural network to function synergistically with an established control scheme (viz. variable structure control) and is significantly different from the popularly discussed approaches for using neural networks for controller designs which mainly attempt to provide a neural network alternative to well established control schemes.