A new methodology to improve the performance of the controller Predictive Functional Control (PFC) in the presence of modelling errors is described. The key idea is to use the available measurements of process internal variables to improve the model predictions without introducing new tuning parametres. Although this approach has been implemented for a particular predictive controller(PFC), its principles can be applied to improve the performance of other predictive controllers because one of most important factors which influence on the Model Based Predictive Control (MBPC) performance is the model prediction accuracy. Finally, some simulated examples are presented.