Treadmill walking endurance is the major outcome measure for intervention studies on patients with intermittent claudication. The measurement properties of this test have not, however, been well documented. We assessed the test-retest reliability of treadmill walking endurance in 10 normal subjects with a mean (+/- 1SD) age of 27.5 (+/- 3.3) years. On four occasions, at intervals of at least one week, each subject walked on a motorized treadmill to the limits of his or her capabilities. With the standard protocol, subjects began at a very low exercise intensity (0.25 m/sec, 0% grade), which increased in either speed or grade at ninety-second intervals through a maximum of fifteen stages to 2.75 m/sec and 20% grade. On the final occasion, 5 of the subjects were tested twice with a fifteen-minute interval, using a truncated protocol starting at stage five (1.25 m/sec, 0% grade). Within-subject standard deviations were low, averaging twenty-three seconds (2.1%). Reliability was very high for the first three tests with the standard protocol (r greater than or equal to 0.98, p less than 0.0001) and for the 5 subjects tested and retested with the short interval and truncated protocol (r = 0.99, p less than 0.0002), while the correlation between the standard protocol and the truncated protocol was somewhat lower (r greater than 0.81, p less than 0.01). For young normal subjects the measurement of walking endurance with an incremental treadmill test is highly reliable, but the limits of reliability need to be considered when conclusions are drawn from such measures.