High retest reliability is desirable in tests used to monitor athletic performance, but the reliability of many popular tests has not been established. The aim of this study was to determine the reliability of performance of a 2000-m time-trial lasting approximately 7 min performed on a Concept II rowing ergometer. Eight well-trained rowers (peak oxygen uptake 61+/-5 ml x kg(-1) x min(-1); mean +/- standard deviation) performed the time-trials on three occasions at 3-day intervals. Mean power (313+/-38 W in trial 1) improved by 2.3% (95% confidence interval 0.1 to 4.5%) in trial 2 and by a further 0.9% (-1.4 to 3.3%) in trial 3. The variability of performance for individual rowers expressed as a coefficient of variation for mean power was 2.0% (1.3 to 3.1%), and the retest correlation was 0.96 (0.87 to 0.99). Variability and changes in performance expressed as time to complete the test were approximately one-third those of mean power, apparently because simulated velocity is proportional to the cube root of power on this ergometer. Such high reliability makes this combination of ergometer, athlete and test protocol very suitable for monitoring rowing performance and for investigating factors that affect performance in short, high-intensity endurance events.