Glutamic acid decarboxylase (GAD) in pancreatic beta cells is an autoantigen in insulin-dependent diabetes (IDD). We measured immunity to GAD in 31 first-degree relatives of IDD patients judged to be at risk of developing IDD themselves because of the presence of islet-cell antibodies. We found that in most of the subjects GAD autoimmunity was either predominantly humoral or predominantly cellular. High concentrations of circulating autoantibodies that precipitate native GAD activity were associated with low proliferation of peripheral-blood T cells to recombinant GAD; conversely, low concentrations of autoantibody to GAD were associated with high T-cell proliferation to GAD. Although T-cell proliferation was measured in the presence of autologous serum, GAD autoantibodies did not have a blocking effect in vitro. This dichotomy of the immune response to GAD defined heterogeneity within at-risk relatives and could have prognostic importance. We postulate that, if GAD is a pathogenetic autoantigen, sensitisation to beta-cell GAD is more likely to lead to IDD when the immune response deviates towards the expansion of autoreactive T cells rather than towards generation of autoantibodies. This idea is consistent with evidence that beta-cell destruction is mediated by T cells and that high concentrations of GAD antibodies are associated with slower progression to clinical disease.