Although the decoding rules have been largely elucidated, the physical-chemical reasons for the “correctness” of codon:anticodon duplexes have never been clear. In this work, on the basis of the available data, we propose that the correct codon:anticodon duplexes are those whose formation and interaction with the ribosomal decoding center are not accompanied by uncompensated losses of hydrogen and ionic bonds. Other factors such as proofreading, base–base stacking and aminoacyl–tRNA concentration contribute to the efficiency and accuracy of aminoacyl–tRNA selection, and certainly these factors are important; but we suggest that analyses of hydrogen and ionic bonding alone provides a robust first-order approximation of decoding accuracy. Thus our model can simplify predictions about decoding accuracy and error. The model can be refined with data, but is already powerful enough to explain all of the available data on decoding accuracy. Here we predict which duplexes should be considered correct, which duplexes are responsible for virtually all misreading, and we suggest an evolutionary scheme that gave rise to the mixed boxes of the genetic code.