We present an adaptive feedback linearizing control scheme for excitation control and power system stabilization. The power system is a synchronous generator which is first modeled as an input-output nonlinear discrete-time system approximated by two neural networks. Then, the controller is synthesized to adaptively compute an appropriate feedback linearizing control law at each sampling instant using estimates provided by the neural system model. This formulation simplifies the problem to that of designing a linear pole-placement controller which is itself not a neural network but is adaptive in the sense that the neural estimator adapts itself online. Additionally, the requirement for exact knowledge of the system dynamics, full state measurement, as well as other difficulties associated with feedback linearizing control for power systems are avoided in this approach. Simulations demonstrate its application to a high-order single-machine system under various conditions.