Twenty-eight brachial arteriovenous fistulae (AVF) flows were assessed by the Stewart and Hamilton method by bolus dye injection. These measurements were divided in two groups: a first group with dye injection into the AVF artery and a second group with dye injection into the efferent vessel of the AVF in close proximity. The increase and the decrease of dye concentration were regular and the circulation occurred very late in both groups. Reproducibility was assessed by the usual index: the mean of the differences between two successive measurements of each series related to the first of these two and expressed as a percentage, m(Qn - Qn - 1)/Qn%. In the two groups, the reproducibility index was at 10.1%, similar to the index applied to Grimby's results, measuring successive cardiac output by dye bolus injection at 9.8%. Theoretical criteria of validity of the Stewart and Hamilton method were checked for all measurements. Even when the duration of the measurement was very short, arteriovenous flow fulfilled the criteria of validity in the same way as cardiac output. Two AVF flows were measured successively at both injection sites with no difference between the obtained values. The same reproducibility of the efferent vessel injection site group suggests that an arterial puncture is not necessary for a correct assessment of AFV flow.