The mapping of molecular surfaces is of particular interest to dye chemists for numerous reasons, none more so than the prediction of dye-substrate binding. Self-organising neural networks have been used to map the hydrogen bonding, electrostatic and hydrophobic 3D molecular surface potentials of a series of dyes. The results indicate that the hydrogen bonding potential, the molecular electrostatic potential and their combination are useful in classifying the dyes and that the hydrogen bonding potential is a useful molecular descriptor of substantivity.