We designed a randomized controlled study to evaluate the benefit of upper-limb exercise training, alone and in combination with walking training, in patients with severe CAO. In an outpatient department supervised by a physiotherapist, we evaluated 28 patients with severe stable CAO (FEV1, 32 percent of predicted). Patients were randomly allocated to either a control (eight), upper-limb (six), lower-limb (seven), or combined (seven) exercise group. The upper-limb group trained with a circuit of upper-limb exercises, the lower-limb group by walking, and the combined group with both. Exercise was for one hour three times per week for eight weeks. Assessment before and after training included pulmonary function, mouth pressures, respiratory muscle endurance, maximal bicycle exercise test, maximal and submaximal arm ergometer, six-minute walking distance, and a scale of well-being (Bandura scale). Twenty-six patients completed the program. There was a significant improvement (Wilcoxon rank sum test) in the following: six-minute walking distance in the lower-limb (p less than 0.005) and combined (p less than 0.003) groups; arm ergometer in the upper-limb (p less than 0.005) and combined (p less than 0.04) groups; and the scale of well-being in the combined (p less than 0.005) group. There was no significant change in any other parameter measured. We conclude that exercise training improves exercise performance in severe CAO, that the training is specific for the muscle group trained, and that upper-limb exercises should be included in training programs for these patients.