Exercise-associated oxygen desaturation in severe chronic obstructive pulmonary disease (COPD) remains a topic of unknown prognostic and clinical relevance and of unknown therapeutic importance. Furthermore, exercise limitation in these patients is due to multifactorial interaction of respiratory, cardiac, circulatory and ‘peripheral’ mechanisms. For the evaluation of the role of the oxygen-dependent mechanisms of exercise limitation we studied 14 stable patients with severe COPD (FEV1 32% pred., FEV1/VC 41%, thoracic gas volume of 180% pred.) and mean Pa O264 mm Hg by a stage 1 cycle exercise test during breathing room air (FiO2 0.21) and hyperoxic (FiO2 0.35) air. The measurements were performed by an open system using a low dead space valve (55 ml) measuring real-time breath by breath oxygen consumption (VO2) via a differential oxygen sensor, expired carbon dioxide (VCO2), ventilation (VE), oxygen pulse and oxygen saturation in both conditions. With FiO2 0.21 the mean exercise limitation was restricted to 37 ( ± 14) W and the VO2 to 11.7 ( ± 2.4) ml/kg/min with peak oxygen desaturation of 86.4 ( ± 3.2). With FiO2 0.35 the patients achieved a significant increase of workload to 55 ( ± 12.4) W and of VO2to 14.9 ( ± 2.9) ml/kg/min. Despite the complete abolition of exercise desaturation by FIO2 0.35 the mean peak ventilation of 261/ min ( ± 5.1) vs. 27 1/min ( ± 4.1) remained unaltered and therefore the restriction of the ventilatory pump limited the exercise ability. The effects of FiO2 0.35 occurred in these patients at submaximum exercise and may be related to reduction of ventilatory cost of energy and delayed onset of metabolic acidosis. We conclude that oxygen treatment in borderline hypoxemic COPD patients who desaturate in exercise is of substantial benefit because oxygen may diminish the generally excessive ventilatory cost of energy at submaximum exercise.