Microbial communities are found in wide array of diverse environments, ranging from the human skin to built freshwater lakes. The species-abundance curve of these communities is usually highly skewed and typical of communities where the diversity mostly comes from a lot of rare species but a few abundant ones contribute the bulk of the individuals. Several comm unity assembly processes have been suggested: they generally assume that local communities are assembled by sampling in some way from a meta pool of species. Neutral processes constitute a useful null model to test observed community structures against exp ected ones. They assume that both species and samples from different environments are exchangeable. Here we use an approach based on the local specificity of a species to its environment to explore this assembly process within a set of communities coming from closely related environments. Our analyses reveal a non-random assembly process, with locally abundant species much more specific to their environment than expected under a neutral model, either using or not overall abundances. A more sophisticated model that accounts for environmental filtering and environment-specific carrying capacities to bias the sampling probabilities achieves a better fit to the data. This underlines the need to incorporate ecological processes when studying assembly processes. At the same time, it suggests that abundant species are the cornerstone of differences between conditions and that shallow sampling efforts are sufficient to identify differences between communities, in accordance with numerous empirical findings.