Abstract The simulation of the operation of an industrial sugar crystallization unit under a generalized predictive controller is reported. The feed flow rate of sugar liquor or syrup and the vacuum pressure have been selected as input control variables. The aim is to force the operation into following optimal profiles of brix and temperature. The state vector in the dynamic model includes temperature, dissolved sucrose and dissolved impurities in liquor, mass of sucrose crystals and the moments of order zero to five in the characterization of crystal size distribution. Main intensive variables such as brix, supersaturation, crystal content and masscuite consistency are calculated from the set of state variables. The control algorithm first predicts the plant output over several steps. It then determines the future control inputs to the plant in order to minimize a quadratic function of the future errors and control, assuming that after some periods nul increments for the control vector. A linear discrete model with time-varying parameters was adopted as a representation of the complex dynamics of the crystallizer. The model parameters are estimated by a constant trace algorithm. Several modifications like the U/D factorization, the normalisation and an information measure (to detect when the information carried out by data is poor) have been introduced to improve the robustness of the parameter estimator towards round-off errors, unmodelled dynamics etc. The dynamic model gives an accurate representation of the industrial behaviour of this batch crystallizer. Control simulations show an improved performance of the proposed control algorithm over classical alternatives (P.I.D.), hence suggesting its industrial implementation.