The complexes of human neutrophil elastase with a series of 40 N3-substituted trifluoromethylketone-based pyridone inhibitors have been modelled. The series spans three orders of magnitude in inhibition constants despite the fact that it was originally developed in an attempt to improve the oral activity of a lead compound. Ligand-receptor interaction energies calculated using molecular mechanics did not correlate well with the experimental activities. A good correlation with activity was found, however, when a COMBINE analysis of the same data was carried out, which allowed a quantitative interpretation of the modelled complexes. The essence of this method is to partition the ligand-receptor interaction energies into individual residue-based van der Waals and electrostatic contributions, and to subject the resulting energy matrix to partial least squares analysis. Incorporation of two additional descriptors representing the electrostatic energy contributions to the partial desolvation of both the receptor and the ligands improved the QSAR model, as did the replacement of the distance-dependent electrostatic contributions with solvent-screened electrostatic interactions calculated by numerically solving the Poisson-Boltzmann equation. The model was validated both internally (cross-validation) and externally, using a set of twelve 6-phenyl-pyridopyrimidine analogs. The analysis reveals the subtle interplay of binding forces which occurs within the enzyme active site and provides objective information that can be interpreted in the light of the receptor structure. This information, gained from a series of real compounds, can be easily translated into 3D real or virtual database queries in the search for more active derivatives.