The delineation of bacterial species is presently done by direct DNA-DNA relatedness studies of whole genomes. It would be helpful to obtain the same genomically based delineation by indirect methods provided that descriptions of individual genome composition of bacterial genomes are obtained and included in species description. The amplified fragment length polymorphism (AFLP) technique could provide the necessary data if the nucleotides involved in restriction and amplification are fundamental to the description of genomic divergences. First, in order to verify that AFLP analysis permits a realistic exploration of bacterial genome composition, we showed the strong correspondence between predicted and experimental AFLP data by using Agrobacterium strain C58 as a model system. Second, we propose a method for determining current genome mispairing and evolutionary genome divergences between pairs of bacteria based upon the arbitrary sampling of genomes with AFLP. The measure of current genome mispairing was validated by comparison with DNA-DNA relatedness data, which are correlated to base mispairing. The evolutionary genome divergence is the estimated rate of nucleotide substitutions that have occurred since strains diverged from a common ancestor. Current genome mispairing and evolutionary genome divergence were used to compare members of Agrobacterium spp. used as a model of closely related genomic species. A strong and highly significant correlation was found between calculated genome mispairing and DNA-DNA relatedness values within genomic species. The canonical 70% DNA-DNA hybridization value used to delineate genomic species was found to correspond to the range from 13% to 13.6% current genome mispairing. These values correspond to 0.097 and 0.104 nucleotide substitutions per site, respectively. In addition, experimental data showed that the large Ti and cryptic plasmids of Agrobacterium had little effect upon the estimation of genome divergence. The evolutionary genome divergence was used for phylogenetic inferences. Data showed that members of the same genomic species clustered consistently as supported by bootstrap resampling. Based upon these results, we propose that the genomic delineation of bacterial species could be based in the future to phylogenetic groups supported by bootstraps, and genome descriptions of individual strains obtained by AFLP recorded in accessible data banks, to eventually replace DNA-DNA hybridization studies.