Lack of assay standardization has precluded cross-study comparison of insulin levels. We exchanged blood samples between the San Antonio Heart and Framingham Offspring Studies to compare insulin measurements. Two randomly selected specimens were chosen for each non-diabetic man and woman in each of the bottom four quintiles and top two deciles of the originally assayed fasting and 2-hour post-challenge insulin distributions: 48 plasma samples from Framingham, and after further stratification by ethnicity, 96 serum samples from San Antonio. Total immunoreactive insulin was originally measured in both studies; we repeated the identical assay on exchanged samples. Repeat assays were performed a mean (SD) of 7.0 (0.8) years after collection in the Framingham study and 4.6 (1.1) years in the San Antonio study. Repeat insulin levels were highly correlated with original levels for both San Antonio samples repeated in Framingham (Pearson r=0.923) and for Framingham samples repeated in San Antonio (r=0.959). Original and repeat San Antonio serum insulin levels were similar (mean fasting and 2-hour combined original level 154 pmol/l vs. 142 pmol/l on repeat in Framingham). Framingham plasma insulin levels repeated in San Antonio were substantially lower than original levels (120 pmol/l vs. 336 pmol/l), as were an additional 12 samples repeat assayed in Framingham (93 pmol/l vs. 320 pmol/l). Repeat rank ordering in both studies was excellent: over 90% of subjects originally classified as hyperinsulinemic (top tertile of the combined distribution) were again classified as hyperinsulinemic upon repeat assay. We conclude that sample exchange for insulin measurement is simple and feasible. Original and repeat insulin levels are highly correlated; subjects originally classified as hyperinsulinemic remain so classified upon repeat assay. Associated regression curves can be used to calibrate insulin levels to a common reference standard, allowing epidemiology studies to compare levels of insulin and associated risk factors.