Interactions between flavour compounds and beta-lactoglobulin (BLG) have been the subject of several studies, but there are no unanimous binding site explanations. In our laboratory, interactions between BLG, and two flavour compounds, beta-ionone and gamma-decalactone, were studied by 2D-NMR spectroscopy. It appears that several amino acids affected by binding of gamma-decalactone are buried in the central cavity, whereas binding of beta-ionone affects amino acids located in a groove near the outer surface of the protein. 2D/3D-QSAR studies were performed using QSAR+ module of Cerius2 and Catalyst. The QSAR equation provided by Cerius involves three molecular descriptors: AlogP98 and two topological connectivity indices (CHI-0 and CHI-1). This model takes into account hydrophobicity and molecular shape more than molecular volume. A relatively flat (e.g. a circle) and elongated (non-branched) shape appears to be able to increase the affinity for BLG. In this way, affinity appears to be strongly related to London dispersive forces. It adequately satisfies internal and external validation and allows a signicant, but not an accurate, prediction of binding of aroma compounds to BLG. The commercially available software Catalyst focuses the modelling on the molecular behaviour of a ligand interacting with a receptor from the point of view of the receptor, but using only information from the ligand. In this way, it appears to be appropriate to identify binding sub-sites on BLG. The 3D-QSAR models generation runs succeeded in providing signicant models which precisely estimated the affinities of sub-sets of compounds. A hydrogen bond acceptor and at least two hydrophobic features constitute the best models. Some model allows explaining the abnormal values of affinity constants of any compounds as -terpineol and highlights the importance of hydrogen bonding. Thus, Catalyst confirms the existence of at least two binding sites on the BLG.