There have been several suggested explanations for the presence of noncoding intervening sequences in many eukaryotic structural genes. They may be examples of 'selfish DNA'1,2, conferring little phenotypic advantage, or they may have some importance in gene expression and/or evolution. It has been suggested that each exon (coding sequence) may represent a structural or functional unit of the encoded protein3,4, for which there is good evidence in the case of immunoglobulin7 and haemoglobin8,9 genes. Exon modification, duplication and recombination may thus be general mechanisms for the rapid evolution of eukaryotic structural genes. In many cases, however, it is not apparent that an exon corresponds to some specific feature of the encoded protein 10,11. We describe here evidence that intron-exon junctions usually map to amino acid residues located at the protein surface, suggesting a restriction on the permitted positions of introns within a gene.