A model-reference predictive control using recurrent neural network is presented for a class of nonlinear industrial processes. The neural control law is developed to minimize a cost function based on the predictive performance criterion and model reference scheme. A real-time adaptive control algorithm, including a neural predictor and model-reference neural predictive controller, is proposed. The adaptive learning rates for both the neural predictor and controller are chosen based on Lyapunov stability theory. Numerical simulations reveal that the proposed control method gives satisfactory tracking and disturbance rejection performance for two illustrate nonlinear discrete time systems. Experimental results for the temperature control of a variable-frequency oil-cooling machine have shown the efficacy of the proposed controller under the condition of set-points changes and external disturbances.