HLA-DR4 is associated with insulin-dependent diabetes mellitus (IDDM) in many populations. Many recent studies suggest that the DR4 effect is really due to DQ3.2, an allele of the nearby DQB1 locus. We used T cell clones, MAb, and allele-specific oligonucleotides to test IDDM and control subjects for DR4 subtypes (Dw4, Dw10, Dw13, and Dw14) and for DR4-associated DQB1 alleles (DQ3.1 and DQ3.2). We find that (a) IDDM is approximately equally associated with alleles of the DRB1 locus (Dw4 and Dw10, combined relative risk, RR = 6.4) and the DQB1 locus (DQ3.2, RR = 5.9); and (b) there is significant interaction, in a statistical sense, between these DR and DQ alleles in IDDM. The only IDDM-associated DR4 haplotypes were those carrying the IDDM-associated alleles at both loci (RR = 12.1); haplotypes with Dw4 or 10 but not DQ3.2, or vice versa, had a RR less than 1. Alternative explanations include: (a) that susceptibility requires specific allelic products of both DR and DQ loci; (b) that the combination of certain DR and DQ alleles marks haplotypes with the true susceptibility allele at a third locus; or (c) that Dw4 and 10 mark haplotypes with an allele at another locus that interacts with DQ3.2. As discussed, this third locus is unlikely to be DQA1 (DQ alpha). The data thus are not easily reconciled with an exclusive effect of HLA-DQ. This information increases our ability to predict IDDM by genetic typing: in the population studied, heterozygotes DR3/[DQ3.2, Dw4] or DR3/[DQ3.2, Dw10] had a relative risk of 38.0 and an absolute risk of 1 in 15.