This study was designed to investigate the reproducibility and clinical relevance of several lung function and exercise test indices in a sample of patients with stable severe chronic obstructive pulmonary disease (COPD). Twenty subjects (ages 67.8 +/- 2.0 years, forced expiratory volume in 1s, [FEV1] 39.7 +/- 2.8% predicted) receiving conventional medical therapy and pulmonary rehabilitation were tested 4 times at 1 month intervals. Testing procedures included lung function (inspiratory vital capacity [IVC], FEV1, plethysmographic functional residual capacity [FRC], specific conductance of the airways (sGaw), single breath transfer factor divided by the alveolar volume [TL/VA]); incremental, progressive, symptom-limited, cycle exercise (maximum work load [Wmax], maximum heart rate [HRmax], maximum ventilation [VEmax], maximum oxygen uptake [VO2max]); and 2 modes of submaximum exercise (12 min walking test [12 MWD] and endurance cycle test). The mean of the absolute value of the individual patient, session-to-session, variation was found to be 0.131 for FEV1, 102 ml/min for VO2max. The within-subject variability was the smallest for HRmax and IVC (mean intrasubject coefficient of variation, [CV intra] 5.0 and 6.5%) and the greatest for TL/VA, the work performed during the endurance cycle test (EW) and sGaw (CV intra 16.5, 19.4, and 22.7%), while it was reasonably low (8.1-10.2%) for all the other variables studied. Calculation of the F ratio of the intersubject variance to the residual (total minus intersubject) variance, interpreted as a signal-to-noise, ratio, yielded the following, in decreasing order: TL/VA, EW, VEmax, VO2max, IVC, FEV1, HRmax, Wmax, sGaw, 12 MWD, FRC. If we assume that a useful variable should combine a low within-subject variability (CV intra less than or equal to 10%) with a high signal-to-noise ratio, we conclude that, among all the variables studied, IVC, FEV1, VEmax, and VO2max are those with the greatest clinical potential for functional assessment in patients with COPD.