Three QSPR models for the prediction of the critical temperature, the critical pressure and the normal boiling point of organic compounds were developed. The chemical structures were selected from the DECHEMA database among those with experimentally available properties and included different chemical groups: single ring, fused rings, halogens, -OH, -COOH, -COOR, -CON, -CN, -NH2 and -NO2, etc. The AMPAC software was used to model the molecules using the AM1 Hamiltonian. The CODESSA software was used to calculate the molecular descriptors and the Heuristic method was chosen to find the best multi-linear relation between the property of interest and the most significant set of descriptors. The final QSPR models showed a significantly higher accuracy with respect to the best available group-contribution method. Comparable results were obtained with respect to other QSPR models despite the different composition of the database, confirming the versatility and robustness of the QSPR method. (c) 2007 Elsevier B.V. All rights reserved.