With a multitude of reaction pathways, poly (ethylene‐terephthalate) (PET) polymerization of industrial practice is complex, and the quality of PET is normally described in terms of several experimentally measured indices. In this paper, parameters estimation of industrial PET reactors is presented as a multi‐objective problem to make the mathematic model consistent with the actual industrial process. Considering the interrelation among parameters and the failure of general optimization algorithms, a new multi‐objective estimation of distribution algorithm is proposed. Kernel density estimation is used to make the new population more suitable for real‐life problems instead of Gaussian model during the evolution of the algorithm. Strategies including selection of kernel width, sampling method and Pareto domination selection are used to explore and exploit the search space more efficiently. With industrial operating data identified in steady state and eliminated from gross error, kinetic parameters are estimated by minimizing carboxyl end group concentration and degree of polymerization simultaneously using the proposed algorithm. The simulation results show that the model with estimated parameters has better predictive performance compared with the experimental parameters. Copyright © 2011 Curtin University of Technology and John Wiley & Sons, Ltd.