HLA-associated relative risks of type 1 (insulin-dependent) diabetes mellitus were analysed in population-based Swedish patients and controls aged 0-34 years. The age dependence of HLA-associated relative risks was assessed by likelihood ratio tests of regression parameters in separate logistic regression models for each HLA category. The analyses demonstrated an attenuation with increasing age at onset in the relative risk for the positively associated DQB1*0201-A1*0502/B1*0302-A1*0301 (DQ2/8) genotype (P = 0.02) and the negatively associated DQB1*0602-A1*0102 (DQ6.2) haplotype (P = 0.004). At birth, DQ6.2-positive individuals had an estimated relative risk of 0.03, but this increased to 1.1 at age 35 years. Relative risks for individuals with DQ genotype 8/8 or 8/X or DQ genotype 2/2 or 2/X, where X is any DQ haplotype other than 2, 8 or 6.2, were not significantly age-dependent. An exploratory analysis of DQ haplotypes other than 2, 8 and 6.2 suggested that the risk of type 1 diabetes increases with age for DQB1*0604-A1*0102 (DQ6.4) and that the peak risk for the negatively associated DQB1*0301-A1*0501 haplotype is at age 18 years. There was also weak evidence that the risk for DQB1*0303-A1*0301 (DQ9), which has a positive association in the Japanese population, may decrease with age. We speculate that HLA-DQ alleles have a significant effect on the rate of beta cell destruction, which is accelerated in DQ2/8-positive individuals and inhibited, but not completely blocked, in DQ6.2-positive individuals.