Proinsulin autoantibodies (PAA) are a potential alternative to insulin autoantibodies (IAA) for the prediction of type 1 diabetes. We measured the prevalence of PAA and IAA in 179 patients with newly diagnosed type 1 diabetes and 1028 schoolchildren, and compared the potential of these antibodies for disease prediction. Antibodies were measured using a novel microassay in which screening for binding of radio-labelled antigen was followed by competitive displacement. Thresholds for both antigen binding and competitive displacement were selected so that equivalent numbers of patients with type 1 diabetes were identified by the proinsulin and insulin antibody assays. Similar numbers of schoolchildren were found to have insulin and proinsulin binding above the screening threshold but, after competitive displacement, the prevalence of PAA (3.7%) was more than twice that of IAA (1.4%) with 66% sensitivity. Both PAA and IAA were present in 113 (63%) of patients, as against 0.8% of schoolchildren, implying that individuals with both antibodies are at particularly high risk of disease. The majority of samples with IAA or PAA were displaced with both insulin and proinsulin, suggesting that the main epitopes recognized by these antibodies are on the insulin molecule. We conclude that IAA are more specific than PAA for the prediction of type 1 diabetes.