The species pool for a site is defined as that set of species which have a non-zero probability of maintaining viable populations under the prevailing environmental conditions. it may contain many more species than are actually present in the community of the site. The science of community assembly attempts to understand how particular communities arise from the welter of possible species combinations. The assembly of a grassland plant community from the local species pool was examined in a phylogenetically corrected trait-based study. Competition theory suggests coexisting species should be less similar than expected by chance, whilst environmental sorting theory suggests they should be more similar. This work suggests that, at the whole community scale, species tend to be more similar and that their likelihood of occurrence in communities can to an extent be predicted from their traits. Experimental studies revealed a complicated picture. Species naturally occurring in the community did not show convincing signs of outperforming their absent congeners. Community composition appears to depend in considerable measure upon chance events such as seed dispersal coinciding with the availability of vacant microsites in the community, rather than just a sorting process in which the best suited species are invariably present. These findings suggest that it modelling community assembly is possible, but that it is unlikely ever to be an exact science because it is influenced to a large extent by unpredictable events.