The human leukocyte antigen (HLA)-DQA1*0102/DQB1*0602/DRB1*1501 (DR2) haplotype confers strong protection from type 1 diabetes. Growing evidence suggests that such protection may be mostly encoded by the DQB1*0602 allele, and we reported that even first degree relatives with islet cell antibodies (ICA) have an extremely low diabetes risk if they carry DQB1*0602. Recently, novel variants of the DQB1*0602 and *0603 alleles were reported in four patients with type 1 diabetes originally typed as DQB1*0602 with conventional techniques. One inference from this observation is that DQB1*0602 may confer absolute protection and may never occur in type 1 diabetes. By this hypothesis, all patients typed as DQB1*0602 positive with conventional techniques should carry one of the above diabetes-permissive variants instead of the protective DQB1*0602. Such variants could also occur in ICA/DQB1*0602-positive relatives, with the implication that their diabetes risk could be significantly higher than previously estimated. We therefore sequenced the DQB1*0602 and DQA1*0102 alleles in all ICA/DQB1*0602-positive relatives (n = 8) previously described and in six rare patients with type 1 diabetes and DQB1*0602. We found that all relatives and patients carry the known DQB1*0602 and DQA1*0102 sequences, and none of them has the mtDNA A3243G mutation associated with late-onset diabetes in ICA-positive individuals. These findings suggest that diabetes-permissive DQB1*0602/3 variants may be very rare. Thus, although the protective effect associated with DQB1*0602 is extremely powerful, it is not absolute. Nonetheless, the development of diabetes in individuals with DQB1*0602 remains extremely unlikely, even in the presence of ICA, as confirmed by our further evaluation of ICA/DQB1*0602-positive relatives, none of whom has yet developed diabetes.