Metacommunity theory employs several models (neutral, sorting, patch dynamics, mass effect) that differ in assumptions and predictions about the dynamics of community composition and specifically about local and regional species extinctions and colonizations. Field tests of the theory are predicated on the ability to differentiate among these models but have treated all species in a community as similar (e.g, equivalent in the neutral model). However, some species are habitat specialists, while other species are habitat generalists. This complication can be turned into an advantage to test the theory because it allows unique predictions (or prediction combinations) to be associated with a specific metacommunity model. We used natural microcosm communities to examine the relationship between the metacommunity models and habitat specialization of species. The study system consists of 49 miniature rock pools on north coast of Jamaica from which community and environmental data were collected over the past 16 years. We used partial Redundancy Analysis to relate the variation in habitat generalists and specialists to environmental, temporal and spatial sets of variables. Spatial variables were only important for generalist species and explained 23-43% of variation in their densities, whereas environmental variables (e.g., pH, oxygen concentration) explained 32-47% of variation in specialists. Thus from the metacommunity perspective, habitat specialists experience a combination of species sorting and mass effects, but habitat generalists experience patch or neutral dynamics. Consequently, testing any or all metacommunity models is likely to fail or yield ambiguous outcome unless species inequalities are explicitly considered.