OBJECTIVE--To assess a non-invasive test for endothelial dysfunction, an important early event in the atherogenic process. METHODS--Using high resolution ultrasound, the accuracy of detecting small changes in vessel diameter was assessed using phantom "arteries", and the same equipment was then used to measure flow mediated dilatation in the brachial artery of 40 healthy adults aged 22-51 years, studied on four occasions; intervals between scans were 1-2 days, 1-2 weeks, and 2-4 months. RESULTS--Differences between pairs of phantom "arteries" with diameters 0.1-0.2 mm apart were correctly estimated in 162 of 264 cases (61%); no measurement by any of four independent observers was > 0.1 mm in error, and the mean error was 0.04 mm. For in vivo scans, the overall coefficient of variation for flow mediated dilatation was 1.8% (1.6% for women, 1.9% for men, P = 0.18). In 34/40 subjects (85%), all values for flow mediated dilatation were within 2.5% of the overall mean for each subject. A nested analysis of variance showed the expected between patient variability, and also significant day to day variation, but little between weeks or months. Using these data to generate power function analyses, we calculated that for individuals, an improvement in flow mediated dilatation of 4-8% is significantly greater than natural variability. In clinical trials, a mean improvement in flow mediated dilatation of at least 2% would usually be required to detect a treatment benefit, with much larger subject numbers needed for a parallel group compared to a crossover trial design. CONCLUSIONS--Vascular responses to endothelium dependent and independent stimuli in systemic arteries can be studied non-invasively in man. Subjects should be studied on at least two occasions before and after any intervention, to optimise the chance of showing a significant effect from any potentially beneficial therapy.