When physical performance is monitored with an ergometer, random error arising from the ergometer combines with biological variation from the subject to limit the precision of estimation of performance changes. We report here the contributions of ergometer error and biological variation to the error of measurement in a performance test with two popular cycle ergometers (air-braked Kingcycle, mobile SRM crankset) and a relatively new inexpensive mobile ergometer (PowerTap hub). Eleven well-trained male cyclists performed a familiarization trial followed by three 5-min time trials within 2 wk on a racing cycle fitted with the SRM and PowerTap and mounted on the Kingcycle. Mean power output in each trial was recorded with all ergometers simultaneously. A novel analysis using mixed modelling of log-transformed mean power provided estimates of the standard error of measurement as a coefficient of variation and its components arising from the ergometer and the cyclists. The usual errors of measurement were: Kingcycle 2.2 %, PowerTap 1.5 %, and SRM 1.6 % (90 % confidence limits +/- 1.3). The components of these errors arising purely from the ergometers and the cyclists were: Kingcycle 1.8 %, PowerTap 0.9 %, SRM 1.1 %, and cyclists 1.2 % (+/- 1.5). Thus, ergometer errors and biological variation made substantial contributions to the usual error of measurement. Use of the best ergometers and of test protocols that reduce biological variation would improve monitoring of the small changes that matter to elite athletes.