Ecologists often classify species into binary groupings such as woodland or non-woodland birds. However, each ecologist may apply a different classification, which might impede progress in ecology and conservation by precluding direct comparison between studies. This study describes and tests a method for deriving empirically-based, ecologically-relevant species groups, using Australian woodland birds as a case study. A Bayesian hierarchical model investigates how vegetation and species traits drive birds' preference for woodland vegetation, characterised by low density trees with an open canopy structure. Birds are then classified according to their affinity to areas with high tree cover and woodland vegetation. Interestingly, no traits are strongly associated with species occurrence in woodland habitats, but occurrence in open country and forests differ depending on dispersal ability and foraging habits. Our results suggest that Australian woodland birds may be united by their avoidance of both sparsely-treed and densely-treed habitat, rather than by shared traits. Classifying species according to our groupings provides results consistent with literature on how woodland birds respond to clearing, grazing and urbanisation. Thus, our model is consistent with current ecological understanding regarding woodland birds; it also provides more nuanced inference across ‘closed-woodland’, ‘open-woodland’, ‘forest’ and ‘open country’ groups. We propose that our modelling approach could be used to classify species for other locations and taxa, providing transparent, ecologically-relevant animal groupings.