The objective of the present study was to determine the variability of the arterio-venous O(2) concentration difference [C(a-v)O(2)] at anaerobic threshold and at peak oxygen uptake (VO(2)) during a progressively increasing cycle ergometer exercise test, with the purpose of assessing the possible error in estimating stroke volume from measurements of VO(2) alone. We sampled mixed venous and systemic arterial blood every 1 min during a progressively increasing cycle ergometer exercise test and measured, in each blood sample, haemoglobin concentration and blood gas data. Ventilation, VO(2) and CO(2) uptake were also measured continuously. We studied 40 patients with normal haemoglobin concentrations and with stable heart failure due to ischaemic or idiopathic cardiomyopathy. Mean values (+/-S.D.) for C(a-v)O(2) were 7.8+/-2.6, 13.0+/-2.4 and 15. 0+/-2.7 ml/100 ml at rest, anaerobic threshold and peak VO(2) respectively. The patients with heart failure were divided into classes according to their peak VO(2). Classes A, B and C contained patients with peak VO(2) values of>20, 15-20 and 10-15 ml.min(-1). kg(-1) respectively. At anaerobic threshold, C(a-v)O(2) was 12.3+/-1. 3, 13.1+/-2.7 and 13.5+/-2.6 ml/100 ml for classes A, B and C respectively (class A significantly different from classes B and C; P<0.05). At peak exercise C(a-v)O(2) was 13.6+/-1.4, 15.6+/-2.5 and 15.4+/-3.2 ml/100 ml for classes A, B and C respectively (class A significantly different from classes B and C; P<0.05). Stroke volume was estimated for each subject using the mean values of the measured C(a-v)O(2) in each functional class and individual values of VO(2) and heart rate using the Fick formulation. The average difference between the stroke volume estimated from mean C(a-v)O(2) and that obtained using the patient's actual C(a-v)O(2) value was 9.2+/-9.7, 1.0+/-8.8 and -0.2+/-6.1 ml at anaerobic threshold, and -1.9+/-11.3, 0.9+/-10.0 and -2.3+/-8.5 ml at peak exercise, in classes A, B and C respectively. Among the various classes, the most precise estimation of stroke volume was observed for class C patients. We conclude that stroke volume during exercise can be estimated with the accuracy needed for most purposes from measurement of VO(2) at the anaerobic threshold and at peak exercise, and from population-estimated mean values for C(a-v)O(2) in heart failure patients.