Flow-mediated vasodilation (FMD) of the brachial artery is used as a marker of cardiovascular disease risk. It is defined as the percentage dilation from the baseline diameter in response to a provoked increase in blood flow. The within-subject variability, crucial in the design of trials with FMD as an endpoint, appears to vary widely between studies. We assessed the analytical and within-subject variability of FMD in healthy subjects and estimated the number of subjects needed to detect various treatment effects in intervention trials and observational studies. FMD was assessed with B-mode high-resolution ultrasound (US). A total of 13 volunteers were measured on six occasions, after they had fasted overnight. Within-subject variability was assessed from all six scans per subject. Analytical variation or reading variation was assessed by reading one scan of each subject twice by one observer. The mean (+/-SD) FMD was 5.60 +/- 2.15 FMD% of the baseline diameter. The within-subject SD was 2.8 FMD%, resulting in a coefficient of variation (CV) of 2.8/5.6 x 100% = 50.3%. The CVs for the baseline and maximum diameter were much smaller: 4.8% (SD 0.193 mm at a mean of 4.060 mm) for the baseline and 5.2% (SD 0.222 mm at a mean of 4.285 mm) for the maximum. The CV for reading variation was 34%. The number of subjects needed to detect a treatment difference of 2 FMD% with a probability of 0.05 and a power of 0.80 would be 31 in a crossover design and 62 per group in a parallel design for comparison of group changes. We conclude that the within-subject variability of FMD is large, about 50% of the mean response. This includes biologic and reading variation. Repeated measurements and repeated readings of recorded measurements are recommended to reduce variability.