For the past twenty years the type 1 diabetes autoantigen glutamic acid decarboxylase (65 kDa isoform; GAD65) has become a prototypic autoantigen, yielding a wealth of immunological and clinical insights. However for most of that period, much of the data could not be placed in a structural context, and relied upon modelling guess-work . The high-resolution crystal structure of GAD65, as well as that of its isoform GAD67, was determined in 2007, providing many insights into the molecular determinants of antigenicity, as well as an atomic positioning of the epitope-mapping data. Despite the two isoforms having the same fold and high sequence identity, it is intruiging that only the 65 kDa isoform functions as an autoantigen. The structures shed much light on this question, revealing striking differences in structure and mobility at the C-terminal domain of the isoforms, which agreed with remarkable accuracy with epitope-mapping data. Furthermore the structures provided an explanation of why two enzymes are required to catalyse the same reaction in mammals, and how this might be linked to their contrasting antigenicities. This review thus focusses on how the GAD system represents a unique testbed for understanding the relationships between molecular structure, function and antigenicity.