Nine selected sera were studied using radioimmunoassay and enzyme linked immunosorbent assay; eight contained insulin antibodies and were from Type 1 (insulin-dependent) diabetic patients, one of whom had antibody-mediated insulin resistance, and one contained insulin-autoantibodies and was from an asymptomatic blood donor. Sera were assayed in serial dilution to assess their suitability for use as reference standards. Dilution curves were non-parallel in radioimmunoassay but were parallel in immunosorbent assay. In all sera, insulin antibodies were readily detected in both assays whereas the low avidity insulin autoantibodies were only detected by immunosorbent assay and not at all by radioimmunoassay, suggesting that the assays respond differently to antibodies of different avidity. Avidity was estimated in liquid phase from the dissociation rate of preformed complexes of antibody and 125-iodinated insulin. When high avidity antibodies are used as a reference in radioimmunoassay, lower avidity antibodies are underestimated and vice versa. In contrast, in immunosorbent assay, any serum could be used as a reference regardless of avidity; furthermore competition experiments comparing the highest avidity insulin antibodies, from the insulin-resistant patient, with the insulin autoantibodies from the asymptomatic blood donor yielded near-superimposable curves. We conclude that radioimmunoassay is selective for high avidity antibodies whereas enzyme linked immunosorbent assay is not; computer modelling of the two assays supports this conclusion. In practice immunosorbent assay can be standardized using a reference serum, whereas experimental findings and mathematical considerations preclude the use of a standard serum in radioimmunoassay.