Recent years have seen considerable advances in ecological understanding of the functional role(s) of biodiversity and the connections between biodiversity, ecosystem function and ecosystem service provision. Functional approaches have become important tools for simplifying biodiversity-ecosystem function relationships, but they also have some obvious weaknesses. In particular, since analyses that use functional groups treat members of a group as ecologically interchangeable, functional groups must be defined at a level that simplifies ecological complexity yet retains key ecological distinctions between groups of species. We developed a data-driven approach to functional group definition and applied it to a case study of 16 species of seed-dispersing Afrotropical waterfowl for which we created seed dispersal functional groups using both a priori categories, as typically done in previous studies, and a hierarchical clustering approach. Relevant functional differences and similarities occur among the waterfowl, particularly in the types of plant family dispersed. We found evidence for at least five functional groups of seed disperser. The different groupings have important implications for both wetland and terrestrial plant dispersal. Our analysis suggests that even for a relatively data-scarce study system, using a data-driven approach to generate functional groups offers a feasible and ecologically rigorous approach and is a useful alternative to simple a priori classification schemes. Our approach is capable of capturing variation across several functional traits and suggests that existing datasets may be useful in exploring variation in biodiversity-ecosystem function relationships. Since functional classification schemes may affect our conclusions about biodiversity, ecosystem function and ecosystem service provision, considerable care should be given to ensuring that functional groups are not defined in such a way as to mask important ecological differences among supposedly similar species.