This work proposes a supervisory control system based on an optimization layer to calculate the optimum pH and dissolved oxygen (DO) set-points for the SISO controller, maintaining the process at stable partial nitrification. Takagi-Sugeno fuzzy multimodels were implemented to estimate ammonium degradation and nitrite accumulation from on-line DO and pH values, and updated using off-line measurements. An activated sludge reactor was operated successfully over 115 consecutive days with the supervisory control system, achieving ammonium degradation and nitrite accumulation values higher than 95% and 80%, respectively. The on-line estimates of the multimodels showed a prediction error of less than 7% at steady state operation, and reflected the tendencies shown in experiment to be caused by changes in pH value and DO concentration. Crown Copyright (c) 2008 Published by Elsevier B.V. All rights reserved.