Insulin autoantibodies (IAAs) are currently the subject of intensive investigation as potential markers for autoimmune insulitis. However, the results of different reports vary widely. In an attempt to elucidate the reasons for the discrepant reports and to initiate standardization procedures, the International Diabetes Workshop (IDW) undertook two studies in which 22 centers worldwide measured IAA in coded samples. The variance in binding signal from the 49 sera in study 1 was considerable, even when results were standardized, but was largely systematic and attributable to basic differences in assay type (liquid phase versus solid phase) and to differences in the ligand used (human vs. nonhuman insulin). In study 2, 5 sera were prepared and presented blindly to compare dilution curves, insulin-species specificity, interference from irrelevant serum proteins, precision, and dose-dependent displaceability. Many assays, both liquid and solid phase, were influenced by marked and unpredictable nonspecific binding, revealed by loss of parallelism between dilution curves in pooled normal serum and buffer, by variable binding signals with different normal sera, and by difficulty in distinguishing human insulin-specific from cross-reactive IAA sera. It was concluded from the experience of both studies that variance could probably be reduced by using a standard curve with derived common units, a single species of ligand, and methodology to minimize the effect of nonspecific binding. Variation related to assay type was probably due to liquid- versus solid-phase systems being differentially more sensitive to certain aspects of antigen-antibody binding; this issue will be addressed in future serum exchanges and workshops.