The primary structures of human and rabbit fl-globin mRNAs are compared. Using as a standard the extent of nucleotide substitutions inferred from the hypervariable amino acid residues of fibrinopeptides A and B, which are thought to change largely by neutral evolution, we show that not all silent mutations in globin mRNA are neutral. The divergence of the sequences is limited in part by the selective usage of synonymous codons. The divergent nucleotides tend to be distributed nonrandomly: in the coding region silent substitutions are most rare in segments that are also deficient in substitutions leading to replacements.Until recently, the molecular evolution of nucleic acids could not be studied with the directness possible for proteins (1), because the information was based largely on reassociation and hybridization data, or on inferences from protein sequences. However, simplified DNA (2, 3) and RNA sequencing techniques are now available, and should provide a rapid increase in our direct knowledge of the primary structure of genes. (11,12) and divided by the time elapsed since divergence of the two species. However, if two pairs of sequences (e.g., /-globin mRNAs and fibrinopeptide mRNAs; see below) are being compared in the same two species, or in species that have diverged at the same time, the percent substitution values can be treated directly as nonlinear substitution rates. This is very useful, because paleontological estimates of the time of divergence are often subject to considerable uncertainty (13,14).Replacement-corrected percent substitution. By definition, in some silent substitution sites some substitutions may lead to amino acid replacement (e.g., U -G in CAU). To correct for possible substitutions leading to replacements, the number of silent sites can be multiplied by a "replacement correction" factor, which is the sum of possible substitutions that would be silent, divided by the sum of all possible substitutions (i.e., 3 times the number of sites).A neutrality standard for mRNA sequences The "neutral mutation-random drift" theory (15-17) implies that some features of macromolecules are relatively unimportant, so that a significant number of neutral mutations are possible and can be fixed in the population by random drift.Neutralists agree that many mutants are deleterious and therefore nonobserved. However, sometimes they consider all or nearly all possible silent mutations as neutral (18,12), which would imply that no appreciable selection is operating on the mRNA itself (e.g., on processing, secondary structure, stability, translatability, etc.