Twenty-three topological indices were calculated for a data set of 667 different chemicals in an effort to further underline their chemical meaning. Several analyses using chemometric methods were performed on this data set. A refined model, obtained by using principal components analysis, was achieved with only ten relevant topological descriptors. The representation of the data set, in terms of the first six principal components, was used to determine a flexible measure of similarity for chemicals; this representation is then used to deal with different chemical problems and to allow chemical interpretations of topological index behaviour.