Classification of species into dietary guilds that consume similar food resources can provide a useful framework for understanding ecosystem function. Traditionally, species were assigned to guilds qualitatively rather than quantitatively, largely due to a lack of detailed data on the diets of many species. More recently, detailed dietary data are collected and analysed quantitatively to produce a classification, but the collection of such data is intensive and usually limits a study to a small number of species. Here, we show that quantitative classifications can be built from qualitative data sourced from natural history texts and expert knowledge for well-known fauna. We collected available data on the presence or absence of nine broad food item categories in the diets of 158 species of fish observed on New Zealand's shallow rocky reefs. These data were analysed using multivariate statistical methods pioneered by K. R. Clarke and colleagues for use with site-by-species ecological survey data. In particular, we adapted a taxonomic dissimilarity measure to take into account a natural hierarchy of the food items, so that, for example, the diets of two species that consume different types of invertebrates were considered more similar to each other than those of, say, invertivores vs piscivores, or carnivores vs herbivores. Our analysis produced five distinct dietary guilds: Herbivore (9 species), Omnivore (11), Invertivore (80), Macro-carnivore ( 52) and Piscivore (6). We suggest that invertivores, which made up around half of the species, are likely underrepresented in many studies due to the exclusion of smaller species. The approach of collecting broadly available qualitative data and applying modern multivariate quantitative methods to produce more inclusive dietary classifications may be usefully applied to other well-known taxonomic groups for which detailed data are scarce.