The association between HLA-DR and -DQ and insulin-dependent diabetes mellitus (IDDM) in a defined high-incidence area was analyzed in a total of 58 population-based patients, representing 77% of IDDM patients with age at onset below 16 years, and in 92 unrelated parents in control families without IDDM. HLA haplotypes were confirmed by analyzing first-degree relatives in both groups. Seven different methods were used to analyze risk: (1) odds ratio, (2) absolute risk, (3) haplotype relative risk, (4) transcomplementation relative risk, (5) relative predisposing effects, (6) stratification analysis, and (7) test of predisposing allele on haplotype. DQB1*0302 indicated somewhat higher risk than did DR4, while DR3 had a higher risk than DQB1*0201; however, the 95% confidence intervals of the risk estimates overlapped. The positive association between IDDM and the DQB1*0201-DQA1*0501-DR3 haplotype seems to be due to DR3 or to an unknown linked gene. More important, DQA1*0301 was present among 93% of the patients, and this allele in various transcomplementation combinations with DQB1 alleles showed closer association to IDDM than did any other alleles. The strong negative association of the DQB1*0602 allele also in the presence of either DR4 or DQB1*0302 or both suggests that, in a high-risk population for IDDM, HLA-DQ primarily confers protection, perhaps by induction of tolerance. Consistent with known functions, HLA-DR may primarily confer susceptibility, perhaps by induction of autoreactive T lymphocytes.