The genetic code is established by highly specific aminoacylation reactions catalyzed by a set of 20 ancient enzymes, aminoacyl-tRNA synthetases (AARSs), which attach specific amino acids to their cognate tRNAs in an ATP-dependent manner. Modern AARSs are highly evolved enzymes which adhere to complex recognition and editing mechanisms to achieve specificity for their cognate substrates and avoid errors in protein translation which are detrimental to cell health. Recently identified mutations in AlaRS and GlyRS genes have been linked to certain neuropathies in mice and humans. These disease-causing mutations often target aminoacylation function by reducing discrimination against noncognate substrates. Amino acid discrimination by synthetases occurs at two sites -- one for amino acid activation and aminoacylation (synthetic site) and one for editing misactivated amino acids (editing site). While the synthetic site sieves out bulkier amino acids, it misactivates amino acids with side chains smaller than the cognate one. These in turn are hydrolyzed at the editing site. Paradoxically, while AlaRS activates Gly as well as Ala, it also activates the sterically larger substrate Ser. Crystallographic analysis of a catalytic fragment of AlaRS with small substrates bound shows that the sieve for excluding the noncognate Ser from the active site is broken because of selective pressure to retain ATP binding. Specificity of AlaRS for its cognate tRNA depends on a single conserved and position-specific G:U base pair in the tRNA^Alaacceptor stem. This key identity element shifts in position along the acceptor stems of mitochondrial tRNA^Alas, compared with its canonical position in cytosolic tRNA^Alas. The crystal structure of AlaRS, its docking model with tRNA^Ala, and mutagenesis and functional data show how mitochondrial AlaRS adapted in evolution to a varying G:U position to preserve specificity. This work was supported by NIH grants GM15539 and GM23562.