This study tested the hypothesis that severity of respiratory disability may affect the outcome of pulmonary rehabilitation. In this randomized, controlled study, 126 patients with chronic obstructive pulmonary disease (COPD) were stratified for dyspnoea using the Medical Research Council (MRC) dyspnoea score into MRC3/4 (Moderate) (n=66) and MRC 5 (Severe) dyspnoeic (n=60) groups. The patients were randomly assigned to an eight week programme of either exercise plus education (Exercise group) or education (Control group). Education and exercise programmes for the moderately dyspnoeic patients were carried out in a hospital outpatient setting. Severely dyspnoeic patients were all treated at home. Those in the Exercise group received an individualized training programme. There was a significant improvement in shuttle walking distance in the moderate dyspnoeic group, who received exercise training; baseline (mean+/-SEM) 191+/-22 m, post-rehabilitation 279+/-22 m (p<0.001). There was no improvement in exercise performance in the severely dyspnoeic patients receiving exercise. Neither group of control patients improved. Health status, assessed by the Total Chronic Respiratory Disease Questionnaire score, increased in the moderately dyspnoeic patients receiving exercise from 80+/-18 to 95+/-17 (p<0.0001) after rehabilitation. Much smaller changes were seen in the other three groups. Improvement in exercise performance and health status in patients with chronic obstructive pulmonary disease after an exercise programme depends on the initial degree of dyspnoea.