Mistranslation can follow two events during protein synthesis: production of non-cognate amino acid:transfer RNA (tRNA) pairs by aminoacyl-tRNA synthetases (aaRSs) and inaccurate selection of aminoacyl-tRNAs by the ribosome. Many aaRSs actively edit non-cognate amino acids, but editing mechanisms are not evolutionarily conserved, and their physiological significance remains unclear. To address the connection between aaRSs and mistranslation, the evolutionary divergence of tyrosine editing by phenylalanyl-tRNA synthetase (PheRS) was used as a model. Certain PheRSs are naturally error prone, most notably a Mycoplasma example that displayed a low level of specificity consistent with elevated mistranslation of the proteome. Mycoplasma PheRS was found to lack canonical editing activity, relying instead on discrimination against the non-cognate amino acid by kinetic proofreading. This mechanism of discrimination is inadequate for organisms where translation is more accurate, as Mycoplasma PheRS failed to support Escherichia coli growth. However, minor changes in the defunct editing domain of the Mycoplasma enzyme were sufficient to restore E. coli growth, indicating that translational accuracy is an evolutionarily selectable trait.