If the genetic code arose in an RNA world, present codon assignments may reflect primordial RNA–amino acid affinities. Whether aptamers selected from random pools to bind free amino acids do so using the cognate codons at their binding sites has been controversial. Here we defend and extend our previous analysis of arginine binding sites, and propose a model for the maintenance of codon–amino acid interactions through the evolution of amino acids from ribozyme cofactors into the building blocks of proteins.