A simple neutral model predicts that the ratio of non-synonymous to synonymous fixed differences between species will be the same as the ratio of non-synonymous to synonymous polymorphisms within species. This prediction is tested with existing mitochondrial datasets from 25 animal species. In slightly over half of the studies, the ratio of replacement to silent polymorphisms within species is significantly greater than the ratio of replacement to silent fixed differences between species. These observations are best explained by a substantial number of mildly deleterious amino acid mutations that contribute to heterozygosity but rarely become fixed.