A neural network based predictive controller design algorithm is introduced for non-linear control systems. It is shown that the use of non-linear programming techniques can be avoided by using a set of affine non-linear predictors to predict the output of the non-linear process. The new predictive controller based on this design is both simple and easy to implement in practice. An on-line weight-learning algorithm for 9 neural networks is introduced, and convergence of both the weights and estimation errors is established. Predictive controller design based on the new procedure is illustrated using a growing network example.