Freshwater-acclimated rainbow trout were chronically and non-occlusively cannulated in the dorsal aorta (DA), sinus venosus (SV) and branchial vein (BV), the latter returning the blood perfusing the central venous sinus (CVS) of the gill after being shunted away from the systemic circuit. After recovery, blood samples from these three sites were analysed for haematocrit (Hct) and [Hb]. Branchial venous blood was found to have considerably lower Hct and [Hb] (Hct = 3.5 +/− 3.1%; [Hb] = 1.04 +/− 0.75 g 100 ml-1) than systemic blood (DA: Hct = 24.3 +/− 8.9%, [Hb] = 6.54 +/− 2.90 g 100 ml-1; SV: Hct = 23.1 +/− 8.8%, [Hb] = 6.15 +/− 2.55 g 100 ml-1; means +/− S.D. N = 8), which strongly suggests that plasma skimming occurred at arteriovenous anastomoses and arterioles draining into the CVS. The partitioning of cardiac output, calculated on the basis of the [Hb] data, showed that the systemic flow accounted for 93 +/− 4.6% (N = 7) of the total cardiac output with only 7 +/− 4.7% of cardiac output being diverted into the CVS, thus bypassing the systemic vasculature. Previous results using in vitro experiments are compared with the present data in an evaluation of the usefulness of the isolated perfused gill and head preparations in the experimental analysis of circulatory physiology in fish gills.