In this work the implementation of a model predictive control system in a production plant is described. A detailed kinetic model is combined with correlations, which were derived from a statistical analysis of plant data to obtain a hybrid description of the reactor behavior for a heterogeneously catalyzed gas phase reaction. It is used to calculate key values for the indication of the performance of the catalyst. From the time dependent trends of these values the plant operators can derive specific actions to keep the process yield as high as possible. A material and energy cost improvement was achieved. (c) 2007 Elsevier Ltd. All rights reserved.