The use of a process model within a controller to obtain improved control is described. The approach is to choose the process variables most suitable from a control standpoint for use as measured quantities and, where these are not available as actual measurements, to predict them using the model. Key features of this work are the use of a general nonlinear model and its augmentation by an output-error feedback loop for model parameter adjustment. The first enables a full, rigorous model to be used; if required this can be the same model used for design purposes. The second makes the technique robust to model-process mismatch. The method is illustrated by means of a very simple nonlinear example, and the results of its application to a more complex system are reported.