Glutamic acid decarboxylase (GAD65) has been implicated as a targeted self antigen in the immune destruction of pancreatic beta cells. T cell responses to GAD65 peptides have been detected in both patients with type I diabetes and in the non-obese diabetic (NOD) mouse. To establish which GAD65 epitopes are important in the immunopathogenesis of disease we initially compared T cell responses to GAD65 epitopes in conditions of disease susceptibility and protection. T cell responses to GAD65 peptides were measured in monozygotic twin pairs selected on the basis of disease discordance and T cell recognition of immunogenic regions of GAD65. Peptides of interest were then used to immunize susceptible NOD mice and H2-E transgenic NOD mice which are protected from diabetes. A differential response to the epitope GAD65 521-535 discriminated diabetic from non-diabetic human twins as well as susceptible from protected mice. This epitope as well as GAD 505-519 induces T cell responses despite binding the type I diabetes associated HLA-DQA1*0301/DQB1*0302 product with low affinity. Since DQ-restricted T cell responses are difficult to study in humans, HLA-DQ8 transgenic mice were then used: GAD epitopes 521-535 and 505-519 induced responses in DQ8 transgenic mice and T cell lines were established. Long-term T cell lines against GAD 505-519 were HLA-DQ restricted, and responded to peptide with a strong IFN-gamma and IL-10 response. The findings implicate GAD 521-535 as a possible target peptide in pathogenesis and are compatible with a model whereby self-reactive T cells specific for low-affinity peptide-MHC complexes may escape thymic negative selection.