Insulin autoantibodies (IAA) have been identified in newly diagnosed insulin-dependent diabetes mellitus (IDDM) patients and in individuals at high risk of developing the disease. However, the literature is not in agreement regarding the prevalence, significance, and predictive value of IAA. Previous workshops have shown that certain sera give markedly different results depending upon assay methodology and, therefore, have suggested that these discrepancies may be due to variations in the assay methodologies used: either the fluid phase RIA or the solid phase enzyme-linked immunosorbent assay (ELISA). Sera from controls (n = 61), newly diagnosed IDDM patients (n = 30), healthy subjects who later became diabetic (n = 8), and first degree relatives of diabetic probands (n = 22) were randomly numbered and sent without category identification to 19 RIA and 10 ELISA laboratories. Each laboratory's raw data from the control sera were used to determine the cut-off point for positive (3 SD above the mean of control sera) in the disease relevant categories. RIA and ELISA methods were comparable in obtaining a low frequency of IAA in control sera. However, the laboratories using RIA methods found a much higher percentage of sera to be IAA positive among both newly diagnosed patients and healthy individuals who later developed diabetes than laboratories using ELISA methods (P less than 0.005). In contrast, there was considerable overlap in the percentage of sera from first degree relatives found positive by both assays. These data strongly suggest that IAA measured by RIA methodology are more disease related than those measured by ELISA methods. Consequently, RIA assays or assays proven to perform as well should be used to measure IAA associated with IDDM.