Incremental cardiopulmonary exercise testing work rate ideally increases linearly to the subject's tolerance within approximately 10 min. Widely used treadmill protocols often yield shorter exercise times in debilitated patients. We compared a recently described treadmill protocol featuring linear work rate increase, weight adjustments and a priori exercise tolerance estimates with standard cycle and treadmill protocols. We also compared treadmill and cycle responses to examine mechanisms of oxyhaemoglobin desaturation differences. In total, 16 subjects with chronic obstructive pulmonary disease (COPD; mean+/-sd forced expiratory volume in 1 s of 36.5+/-10.9% predicted) performed incremental exercise using cycle, linear treadmill and modified Bruce protocols. Initial linear treadmill speed and grade yielded oxygen uptake (V'(O(2))) similar to cycle unloaded pedalling; Bruce protocol first stage elicited much higher V'(O(2)). Exercise duration was much shorter in Bruce than in cycle or linear treadmill protocols. At peak exercise, greater desaturation was noted in linear treadmill and Bruce protocols compared with cycle (-8.9+/-4.9 versus -8.5+/-4.7 versus -3.7+/-3.3%; p<0.001); at iso-V'(O(2)) values this difference widened as exercise proceeded. Iso-V'(O(2)) desaturation differences were largely related to higher ventilatory response to cycle than to treadmill exercise. The linear incremental treadmill protocol generates responses similar to cycle ergometry in severe COPD. However, cycle ergometry elicits less desaturation than does ambulation, making the linear treadmill protocol advantageous when evaluating COPD patients.