1. All members of 33 families in which two or more sibs have insulin-dependent diabetes mellitus (IDDM) were HLA-typed. The results strongly support the hypothesis that, closely linked to the HLA region, there is a locus (S) for susceptibility to IDDM. We use Sd for alleles at this locus which confer susceptibility to disease, and Sa for all other alleles. 2. Published prevalence (recurrence) rates for relatives of an IDDM proband, which are essential for the analysis, are reviewed critically, and a new estimate of prevalence at age 30 in the general population is calculated from published data. 3. The analysis allows for greater risk (penetrance) of the genotype SdSd (two doses of Sd) than of SaSd (one dose of Sd) and for recombination between the HLA region and S. With either our own data on 26 affected sib pairs or a much larger sample assembled by pooling with published data, the maximum-likelihood estimate of the two-dose penetrance is 71%, of the one dose penetrance 6.5%. The hypothesis of differential penetrance according to gene dosage provides a significantly better explanation than either simple dominant or simple recessive inheritance. 4. Estimates of the recombination fraction between HLA and S are very sensitive to the proportion of affected sib pairs sharing neither HLA haplotype, which varies considerably among the samples pooled. With the pooled data the recombination fraction is estimated as 3%, but in view of the strong associations with particular HLA alleles seen in population data, this should probably be considered an upper limit.