In this paper, a new control technique for nonlinear control based on hybrid neural modeling is proposed. For neural network training, a variant of the well-known gradient steepest descent method is employed where the learning rate is adapted in each iteration step in order to accelerate the speed of convergence. It is shown that appropriate selection of the learning rate results in stable training in Lyapunov sense. The closed-loop control system consists of two neural networks. The first one is a feedforward neural network that is employed as a predictive hybrid model of the controlled plant. The second network is a neural PID-like controller, which has been pre-trained off-line as an inverse black-box model of the controlled process. To ensure offset-free performance, additional on-line tuning of the neural controller is required, especially in the presence of process uncertainties and time-varying parameters. Advantages of the proposed technique are demonstrated through simulation experiments for a case study, investigating the control of a continuous flow stir-red biochemical reactor. (C) 2003 Elsevier Ltd. All rights reserved.