In prokaryotic taxonomy, a set of criteria is commonly used to delineate species. These criteria are generally based on cohesion at the phylogenetic, phenotypic and genomic levels. One such criterion shown to have promise in the genomic era is average nucleotide identity (ANI), which provides an average measure of similarity across homologous regions shared by a pair of genomes. However, despite the popularity and relative ease of using this metric, ANI has undergone numerous refinements, with variations in genome fragmentation, homologue detection parameters and search algorithms. To test the robustness of a 95–96 % species cut-off range across all the commonly used ANI approaches, seven different methods were used to calculate ANI values for intra- and interspecies datasets representing three classes in the
Proteobacteria
. As a reference point, these methods were all compared to the widely used blast-based ANI (i.e. ANIb as implemented in JSpecies), and regression analyses were performed to investigate the correlation of these methods to ANIb with more than 130000 individual data points. From these analyses, it was clear that ANI methods did not provide consistent results regarding the conspecificity of isolates. Most of the methods investigated did not correlate perfectly with ANIb, particularly between 90 and 100% identity, which includes the proposed species boundary. There was also a difference in the correlation of methods for the different taxon sets. Our study thus suggests that the specific approach employed needs to be considered when ANI is used to delineate prokaryotic species. We furthermore suggest that one would first need to determine an appropriate cut-off value for a specific taxon set, based on the intraspecific diversity of that group, before conclusions on conspecificity of isolates can be made, and that the resulting species hypotheses be confirmed with analyses based on evolutionary history as part of the polyphasic approach to taxonomy.