Relative single base-pair substitution rates in human genes, derived from a collection of > 2,700 point mutations causing human genetic disease, were related to the results of an evolutionary gene/pseudogene comparison. At the mononucleotide level, notable differences between the two datasets were confined to C-to-T and G-to-A transitions, both being rarer in gene/pseudogene alignments than among disease-associated lesions. Relative nearest neighbour-dependent substitution rates were found to be similar in the two datasets, indicating the long-term stability of these parameters during human genome evolution. Allowing for the 5' and 3' nucleotides flanking mutated sites, the primary likelihood of mutation generation could be demonstrated to be biased toward the avoidance of replacements that: (1) change the chemical characteristics of the encoded amino acid residue substantially, and (2) have a high chance of resulting in genetic disease in humans. A similar bias is also reflected in the evolutionary history of human and rodent proteins: amino acid replacements that currently exhibit a high likelihood of coming to clinical attention have been less likely to be accepted during protein evolution.