To examine the reproducibility of cardiopulmonary exercise testing in patients with heart failure, three consecutive tests were performed in 30 such patients. The first test underestimated treadmill exercise time by about 20% when compared with the second and third tests, which were not significantly different. Peak achieved VO2, VCO2 and VE were also less during the first test, but blood pressure, heart rate and respiratory rate responses were similar in the three tests. When cardiopulmonary exercise tests are used to assess functional capacity in either individual patients or groups (as in a therapeutic trial), at least two tests should be performed, as a single test is likely to underestimate exercise capacity.