This paper presents the use of robust stability conditions as indicators of multivariable controller performance. The main idea is to estimate the uncertainty in the process model as reflected in such a condition and assess the need for controller re-tuning. Different descriptions of the uncertainty lead to different stability conditions. The more attractive ones are largely unaffected by control input saturation and can assess the performance of a given controller, even if the controller is not actually used in the loop. Although the accuracy of the assessment depends on the presence of adequate excitation, the excitation requirement is quite modest compared to system identification. The method is implemented on a multivariable benchmark example representing furnace models at different operating points. The simulation results are analyzed to draw conclusions on the consistency of the monitoring approach in identifying poorly tuned controllers. (c) 2007 Elsevier Ltd. All rights reserved.