Ellipsometry as an indirect optical measurement method requires the use of optical modelling which include model parameterization. In practice, there are many ways to select a model and its parameters to fit the experimental data. Very often this fact leads to ad hoc decisions, i.e., based on experience or subjective opinion, instead use of some systematic approaches which provide predictive capability. In this paper we use the Akaike and Bayesian information criteria to perform optical model selection and its best parameterization to fit a particular set of ellipsometric data. We demonstrate that this approach accompanied by post hoc study of the interparameter correlations can significantly enhance optical modelling, in particularly, the process of model selection and data interpretation and improve the characterization of multilayered thin-film structures. (C) 2015 Elsevier B.V. All rights reserved.