Ecological theory and biodiversity conservation have traditionally relied on the number of species recorded at a site, but it is agreed that site richness represents only a portion of the species that can inhabit particular ecological conditions, that is, the habitat-specific species pool. Knowledge of the species pool at different sites enables meaningful comparisons of biodiversity and provides insights into processes of biodiversity formation. Empirical studies, however, are limited due to conceptual and methodological difficulties in determining both the size and composition of the absent part of species pools, the so-called dark diversity. We used >50,000 vegetation plots from 18 types of habitats throughout the Czech Republic, most of which served as a training dataset and 1083 as a subset of test sites. These data were used to compare predicted results from three quantitative methods with those of previously published expert estimates based on species habitat preferences: (1) species co-occurrence based on Beals’ smoothing approach; (2) species ecological requirements, with envelopes around community mean Ellenberg values; and (3) species distribution models, using species environmental niches modeled by Biomod software. Dark diversity estimates were compared at both plot and habitat levels, and each method was applied in different configurations. While there were some differences in the results obtained by different methods, particularly at the plot level, there was a clear convergence, especially at the habitat level. The better convergence at the habitat level reflects less variation in local environmental conditions, whereas variation at the plot level is an effect of each particular method. The co-occurrence agreed closest the expert estimate, followed by the method based on species ecological requirements. We conclude that several analytical methods can estimate species pools of given habitats. However, the strengths and weaknesses of different methods need attention, especially when dark diversity is estimated at the plot level.