This paper presents the development and the industrial implementation of a virtual sensor (soft-sensor) in the polyethylene terephthalate (PET) production process. This soft-sensor, based on a feed-forward artificial neural network (ANN), was primarily used to provide on-line estimates of the PET viscosity, which is necessary for process control purposes. The ANN-based soft-sensor (ANN-SS) was also used for providing redundant measurements of the viscosity that could be compared to the results obtained from the process viscometer. It was shown that the proposed ANN-SS was able to adequately infer the polymer viscosity, in such a way so as this soft-sensor could be used in the real-time process control strategy. The proposed control system has Successfully been applied in servo and regulatory problems, thus allowing an effective and feasible operation of the industrial plant. (C) 2008 Elsevier Ltd. All rights reserved.