Two models for phenanthrene sorption prediction are discussed and checked against experimental data. The first model, based on a two-site Langmuir expression, displays too large uncertainties. The second is based on a modified Langmuir relation and provides a better description of the whole dataset. It includes an experimentally derived relation between the site maximal concentration and chemical parameters of the substrate: TOC amount and polarity index (O +N)/C. The model is tested against sorption isotherms of phenanthrene acquired with six different substrates. Calculated values display satisfying accordance with experimental data. Validity of the model is tested with an isotherm that does not belong to the set of data used for the regression, with good results. Some discrepancies may arise from analytical uncertainty and structural aspects not included within the model. (c) 2006 Elsevier Inc. All rights reserved.