It has been suggested that the differences in synonymous codon use bet
ween mammalian genes within a genome are due to differences in the eff
iciency of DNA mismatch repair. This hypothesis was tested by developi
ng a model of mismatch repair, which was used to predict the expected
relationship between the rate of substitution and G+C content at silen
t sites. It was found that the silent-substitution rate should decline
with increasing G+C content over most of the G+C-content range, if it
is assumed that mismatch repair is G+C biased, an assumption which is
supported by data. This prediction was then tested on a set of 58 pri
mate and artiodactyl genes. There was no evidence of a direct decline
in substitution rate with increasing G+C content, for either twofold-
or fourfold-degenerate sites. It was therefore concluded that variatio
n in the efficiency of mismatch repair is not responsible for the diff
erences in synonymous codon use between mammalian genes. In support of
this conclusion, analysis of the model also showed that the parameter
range over which mismatch repair can explain the differences in synon
ymous codon use between genes is very small.