Results of diversity and community ecology studies strongly depend on sampling depth. Completely surveyed communities follow log-normal distribution, whereas power law functions best describe incompletely censused communities. It is arguable whether the statistics behind those theories can be applied to voluminous next generation sequencing data in microbiology by treating individual DNA sequences as counts of molecular taxonomic units (MOTUs). This study addresses the suitability of species abundance models in three groups of plant-associated fungal communities – phyllosphere, ectomycorrhizal and arbuscular mycorrhizal fungi. We tested the impact of differential treatment of molecular singletons on observed and estimated species richness and species abundance distribution models. The arbuscular mycorrhizal community of 48 MOTUs was exhaustively sampled and followed log-normal distribution. The ectomycorrhizal (153 MOTUs) and phyllosphere (327 MOTUs) communities significantly differed from log-normal distribution. The fungal phyllosphere community in particular was clearly undersampled. This undersampling bias resulted in strong sensitivity to the exclusion of molecular singletons and other rare MOTUs that may represent technical artefacts. The analysis of abundant (core) and rare (satellite) MOTUs clearly identified two species abundance distributions in the phyllosphere data – a log-normal model for the core group and a log-series model for the satellite group. The prominent log-series distribution of satellite phyllosphere fungi highlighted the ecological significance of an infrequent fungal component in the phyllosphere community.