Many modelling and statistical methods have been used to relate soil characteristics to other environmental factors. As mentioned above, relationships between soil colour and chemistry have been found, although this research has generally employed standard statistical techniques. Work that has applied more sophisticated pattern-matching or recognition approaches such as those used in artificial intelligence have been largely restricted in the field of soil colour to crop/soil or weed/ soil discrimination (Aitkenhead et al., 2003; Kavdir, 2004; Marchant and Onyango, 2003). Here we apply neural networks to the mining of a soils database in order to identify and make use of patterns between soil colour and physicochemical composition. The intention is not to investigate the strength of correlation between soil colour and any single physical or chemical parameter, but rather (A) to determine the usefulness of colour, measured using both RGB and CIELab coordinates, as a proxy indicator of several soil characteristics, and (B) to determine how accurately soil colour can be predicted using a range of chemical and physical characteristics simultaneously.