Twenty patients with chronic obstructive pulmonary disease (COPD) walked on the treadmill until symptoms limited further exercise. Ventilation, pulmonary gas exchange, arterial and mixed venous blood gas and haemodynamic variables were measured during steady state. It was not possible to predict from any resting cardiopulmonary variables (including static lung function tests) which patients would develop a decline in arterial PO2. In spite of variable PvO2 values at symptom-limited exercise, a good correlation was obtained between arterial haemoglobin saturation with oxygen and the venous admixture fraction. The directional change in PaO2, however, did not always follow that of the total oxygen transport to the tissues as gauged by either the PvO2 or the efficiency of oxygen delivery (oxygen flux/oxygen consumption) when expressed as a percentage of the value predicted for the level of exertion. This was due to the variable and unpredictable response of the cardiac output on exercise. It is concluded that there is a need to assess each patient individually both at rest and on exercise when evaluating arterial oxygenation and oxygen transport. The pattern of exercise-induced change in these oxygen indices is not predictable from resting data and they may even follow opposite directional trends.