Bacterial peptidoglycan amidases are a large and diverse group of enzymes. During the last few years, genomic sequence information has accumulated to an extent such that lists of proven or predicted peptidoglycan amidases can now be expected to be fairly complete. Moreover, representative crystal structures for most groups of phylogenetically related peptidoglycan amidases have been solved. Here, sequence and structural information is combined with published biochemical findings to demonstrate that (a) peptidoglycan amidases have evolved for almost every bond that occurs in peptidoglycan, (b) there are enzymes that share the fold, yet cleave different bonds and (c) there are enzymes that have entirely different folds and must have evolved independently, and yet cleave the same peptide bond. It is shown that despite these complications, some rules can be deduced from the available biochemical and structural information that can be useful to predict the specificity of hypothetical peptidoglycan hydrolases, for which only sequence information is available.