Modelflow, when applied to non-invasive fingertip pulse pressure recordings, is a poor predictor of cardiac output (Q, litre x min(-1)). The use of constants established from the aortic elastic characteristics, which differ from those of finger arteries, may introduce signal distortions, leading to errors in computing Q. We therefore hypothesized that peripheral recording of pulse pressure profiles undermines the measurement of Q with Modelflow, so we compared Modelflow beat-by-beat Q values obtained simultaneously non-invasively from the finger and invasively from the radial artery at rest and during exercise. Seven subjects (age, 24.0 +/- 2.9 years; weight, 81.2 +/- 12.6 kg) rested, then exercised at 50 and 100 W, carrying a catheter with a pressure head in the left radial artery and the photoplethysmographic cuff of a finger pressure device on the third and fourth fingers of the contralateral hand. Pulse pressure from both devices was recorded simultaneously and stored on a PC for subsequent Q computation. The mean values of systolic, diastolic and mean arterial pressure at rest and exercise steady state were significantly ( P < 0.05) lower from the finger than the intra-arterial catheter. The corresponding mean steady-state Q obtained from the finger (Qporta) was significantly ( P < 0.05) higher than that computed from the intra-arterial recordings (Qpia). The line relating beat-by-beat Qporta and Qpia was y =1.55 x -3.02 ( r2 = 0.640). The bias was 1.44 litre x min(-1) and the precision was 2.84 litre x min(-1). The slope of this line was significantly higher than 1, implying a systematic overestimate of Q by Qporta with respect to Qpia. Consistent with the tested hypothesis, these results demonstrate that pulse pressure profiles from the finger provide inaccurate absolute Q values with respect to the radial artery, and therefore cannot be used without correction with a calibration factor calculated previously by measuring Q with an independent method.