Abstract Aim A positive relationship between assemblage diversity (AD) – equivalent to the biotic version of the environment diversity, ED, method – and species diversity has been reported. This being true, reserve networks with many different assemblages would be expected to represent more species than reserve networks including fewer and less different assemblages. This idea was tested using European species atlas distributions of terrestrial vertebrates and plants. It is asked whether: (1) maximizing AD within one group would represent species diversity of this group better than expected by chance; and (2) maximizing AD within one group would represent species diversity of other groups better than expected by chance. Location Europe. Methods Three ordination techniques (non-metric multidimensional scaling, detrended correspondence analysis and correspondence analysis) are used to summarize patterns of compositional turnover within assemblages. p-Median location-allocation models are then calculated from ordination space to measure the degree of expected species representation within the group being sampled as well as the expected representation within other groups. Results are compared with near-optimal solutions obtained with complementarity-based algorithms and to a null model obtained by simulating selection of areas at random. Matrix correlation analysis was also performed to investigate broad patterns of covariation in compositional turnover of assemblages of species belonging to different taxonomic groups and these values were compared with correlation in species richness scores between groups. Results The AD model did not always represent more species of the group being sampled than expected by chance (P < 0.05). Results were independent of the method and taxonomic group considered. Effectiveness of AD within one group to represent species of other groups varied, but in most cases it was worse than using complementarity-based algorithms as a surrogate strategy. Even when correlations indicated high coincidence between assemblages, taxonomic-based surrogates did not always recover more species than expected by chance (P < 0.05). Main conclusions Results are discussed in the light of two possible explanations: (1) the AD model is based on unrealistic assumptions, namely that species have equal probability of having the centre of their distributions anywhere in ordination space and that species display unimodal, symmetrical, bell-shaped response curves to gradients; (2) particular implementation of methods may be inadequate to summarize useful complementarity among assemblages, especially for restricted-range species. We conclude that both arguments are likely to play a role in explaining results, but that opportunities exist to improve performance of existing surrogate strategies.