From 11 studies, a total of 1,792 Caucasian probands with insulin-dependent diabetes mellitus (IDDM) are analyzed. Antigen genotype frequencies in patients, transmission from affected parents to affected children, and the relative frequencies of HLA-DR3 and -DR4 homozygous patients all indicate that DR3 predisposes in a "recessive"-like and DR4 in a "dominant"-like or "intermediate" fashion, after allowing for the DR3/DR4 synergistic effect. Removal of DR3 and DR4 reveals an overall protective effect of DR2, predisposing effects of DR1 and DRw8, and a slight protective effect of DR5 and a predisposing effect of DRw6. Analysis of affected-parent-to-affected-child data indicates that a subset of DR2 may predispose. The non-DR3, non-DR4 antigens are not independently associated with DR3 and DR4; the largest effect is a deficiency of DR2, followed by excesses of DR1, DRw8, and DRw6, in DR4 individuals, as compared with DR3 individuals. HLA-B locus distributions on patient haplotypes indicate that only subsets of both DR3 and DR4 are predisposing. The presence or absence of Asp at position 57 of the DQ beta gene, recently implicated in IDDM predisposition, is not by itself sufficient to explain the inheritance of IDDM. At a minimum, the distinguishing features of the DR3-associated and DR4-associated predisposition remain to be identified at the molecular level. Risk estimates for sibs of probands are calculated based on an overall sibling risk of 6%; estimates for those sharing two, one, or zero haplotypes are 12.9%, 4.5%, and 1.8%, respectively. Risk estimates subdivided by the DR type of the proband are also calculated, the highest being 19.2% for sibs sharing two haplotypes with a DR3/DR4 proband.