Because species–environment interactions are mediated by phenotypic tradeoffs, the maintenance of ancestral traits in some phylogenetic clades and the emergence of evolutionary novelties in others are likely to limit the types of habitats that species occupy, generating phylogenetic habitat filtering. To test for phylogenetic habitat filtering in woody sapling communities in vegetation patches scattered in southern Brazilian grasslands, I estimated if patches of different sizes encompassed species of different phylogenetic groups. I analyzed patch composition with principal coordinates of phylogenetic structure (PCPS), extracted from a matrix of phylogeny-weighted species composition, and compared these results against net relatedness index (NRI) analyses. NRI analysis revealed that most communities were phylogenetically random, and that patches of different sizes did not differ from each other with respect to NRI. The first four PCPS contained ≅ 91% of total variation in phylogeny-weighted species composition. In the first two PCPS, scores of large patches differed from those of small and medium patches, which did not differ from each other. Large patches were associated with basal plant clades, whereas small patches were mostly related to asterids, and medium patches were phylogenetically diverse. Phylogenetic habitat filtering was detected only by PCPS analysis, possibly because NRI analysis does not take into account the habitat specificity of species. Taking phylogenetic habitat filtering into account in comparative studies likely enhances our capability to understand the ways that plants interact with their environment.