The site of the 'first hit' in the APC tumour suppressor gene determines the type of the 'second hit', both in familial adenomatous polyposis (FAP) and sporadic colorectal tumours. Mutations near codon 1300 are associated with loss of heterozygosity (LOH) of the wild-type allele; other tumours tend to have two protein-truncating mutations. In this study, we have confirmed and refined the LOH-associated region in colorectal FAP: allelic loss in adenomatous polyps tended to occur when the germline mutation lay in the region of the APC gene between the first and second beta-catenin degradation repeats (codons 1285-1378). LOH generally occurred by mitotic recombination, leaving two identical alleles, each encoding a protein with one remaining beta-catenin degradation repeat. For patients with germline mutations that truncated the protein before the first repeat (codon 1264), LOH was very rare and tumours generally acquired a somatic mutation which left two, or less often one, repeats remaining in the protein. In our sample set, patients with germline mutations after the second beta-catenin degradation repeat tended to have undetectable, presumably cryptic, somatic mutations in their polyps. Exceptions to these rules were, however, not uncommon. Although the site of the germline mutation was the strongest determinant of the somatic mutation in FAP tumours and most patients showed no clear tendency to acquire specific types of truncating 'second hit', a minority of patients did have unusual somatic mutation spectra in their polyps. Thus, some individuals may be predisposed to particular types of 'second hit' (for example, frameshift rather than nonsense changes). Overall, disease severity (polyp number) did not vary with individuals' spectrum of somatic APC mutations, providing no clear evidence for modifier genes that influence disease severity in this fashion. Our data are consistent with the hypothesis that there exists an optimal level of beta-catenin signalling in colorectal tumours and that the APC mutation spectrum principally reflects this fact. The association between 'first hits' and 'second hits' at APC is not, however, so strong as to suggest that tumorigenesis only occurs if the genotype is optimum; we suggest 'relaxed' terminology, the 'loose fit' model, to describe this situation.