The beat-by-beat non-invasive assessment of cardiac output (Q litre x min(-1)) based on the arterial pulse pressure analysis called Modelflow can be a very useful tool for quantifying the cardiovascular adjustments occurring in exercising humans. Q was measured in nine young subjects at rest and during steady-state cycling exercise performed at 50, 100, 150 and 200 W by using Modelflow applied to the Portapres non-invasive pulse wave (Q(Modelflow)) and by means of the open-circuit acetylene uptake (Q(C2H2)). Q values were correlated linearly ( r = 0.784), but Bland-Altman analysis revealed that mean Q(Modelflow) - Q(C2H2) difference (bias) was equal to 1.83 litre x min(-1) with an S.D. (precision) of 4.11 litre x min(-1), and 95% limits of agreement were relatively large, i.e. from -6.23 to +9.89 litre x min(-1). Q(Modelflow) values were then multiplied by individual calibrating factors obtained by dividing Q(C2H2) by Q(Modelflow) for each subject measured at 150 W to obtain corrected Q(Modelflow) (Qcorrected) values. Qcorrected values were compared with the corresponding Q(C2H2) values, with values at 150 W ignored. Data were correlated linearly ( r = 0.931) and were not significantly different. The bias and precision were found to be 0.24 litre x min(-1) and 3.48 litre x min(-1) respectively, and 95% limits of agreement ranged from -6.58 to +7.05 litre x min(-1). In conclusion, after correction by an independent method, Modelflow was found to be a reliable and accurate procedure for measuring Q in humans at rest and exercise, and it can be proposed for routine purposes.