Application of the method of antigen genotype frequencies among patients to HLA-DR data pertaining to insulin-dependent diabetes mellitus substantiates the results of Rotter et al. and Thomson that a singlepredisposing- allele model is incompatible with the observed data. The method is also modified to take into account the possibility of blanks (untyped antigens) in the assumed homozygotes. The rejection of all intermediate single-allele models is still obtained. A minimum of two HLA-linked predisposing components are necessary to account for the data. The patterns observed are consistent with a three-allele model, in which the two predisposing alleles interact synergistically (negative complementation). Furthermore, a model in which the DR3-associated predisposition allele is recessive in the absence of the other allele and the DR4-associated predisposing allele is additive (dominant) in the absence of the other is more consistent with the data than a model in which both alleles are recessive or additive in the absence of the other.