The development and application of gray-box identification techniques for modelling the bleaching operation in a Bleached Chemi-Thermo Mechanical Pulp (BCTMP) mill are explained. The process is characterized as a delay dominant recycle process with significant input nonlinearities. The identification was carried out using routine operating data in which the outputs were measured irregularly. The effects of these characteristics and consequent modifications of the system identification techniques are discussed. The resulting models are being used for online prediction and model-based controller design at the mill with satisfactory performance. (c) 2004 Elsevier Ltd. All rights reserved.