The repertoire of known autoantigens is limited to a very small proportion of all human proteins, and the reason why only some proteins become autoantigens is unclear, but is likely associated with structural features. The 65kDa isoform of the enzyme glutamic acid decarboxylase (GAD65) is a major autoantigen in type I diabetes, and in various neurological diseases, whereas the closely related isoform, GAD67, is rarely antigenic. Conformational epitopes of GAD65 have been mapped using human monoclonal antibodies to GAD65 and GAD mutated by GAD65/67 sequence exchanges or point mutations, but these studies have been limited by a lack of structural information. The recent publication of crystal structures for the two isoforms has shown that the N-, C- and middle domains that have been identified previously as likely epitope regions are closely associated within the GAD dimer. Two major epitope regions, ctc1 and ctc2, have been identified in the C-terminal domain of GAD65, that encompass N- and C-terminal residues, and middle and C-terminal residues respectively. These regions are highly flexible compared with the equivalent regions in GAD67, and T cell epitopes have been localized to the same surface region of GAD65. Comparative analysis of these two structurally similar isoforms, GAD65 and GAD67, only one of which is autoantigenic should provide new insights into the provocations to autoimmunity.