The presence or absence of 75 forest-core plant species in 156 forest fragments in Western Belgium was surveyed. We performed a nestedness analysis both at the total species pool level and at the individual species level. Community composition of these plant species seemed to exhibit a highly significant nested subset pattern. We tested the classic hypotheses concerning the ecological mechanisms generating nestedness, i.e. stochastic species extinction and differential colonization. Neither patch area nor patch isolation, however, adequately explained the species nestedness pattern, hence our conclusion that these variables are of little use in predicting the ultimate plant species composition of a forest fragment. We then derived a complex synthetic habitat diversity measure for each forest fragment and ranked the patches accordingly. The resulting plant species occurrence patterns approximated the nestedness pattern very closely. Hence we suggest that nested plant species assemblages result in the first place from nested habitats, and that species diversity in forest fragments is determined primarily by habitat diversity and is predictable on the basis thereof. Nested species subsets are not generated by differences in minimal sustainable population sizes between species but by differences in habitat specialization. We conclude that forest plant species extinctions in Western Belgium mere probably not caused by patch-area-dependent stochastic extinction processes resulting from shrinking forest size (i.e. species relaxation), but in contrast must have their origin in deterministic processes caused by overall habitat deterioration. We also learned from our dataset that a significantly positive species-area relationship and significant species nestedness do not necessary imply that one large forest fragment supports more species than several small patches of the same total size. This result is also associated with habitat diversity and the intrinsic high sensitiveness of forest-core plant species.