Sprint tests are frequently used to evaluate between-subject differences and can provide a valuable insight into performance capacity. The present study determined the reproducibility of peak and mean power output during upper-body sprints. After familiarization 25 men (mean [+/- SD] age 29 [6] years, body mass 82.8 [12.7] kg and height 1.76 [0.05] m) completed 2 20-second upper-body sprint tests using an adapted cycle ergometer. Mean (+/- SD) values of all power (uncorrected and corrected) measurements achieved during the 2 tests were checked for systematic bias using separate paired t-tests. Test-retest reproducibility was examined using coefficients of variation and single-measure intraclass correlation coefficients, as well as an assessment of the typical (random) error and the 95% limits of agreement. The value of corrected peak power (628 [167] W) was higher (p < 0.05) compared with the uncorrected value (509 [109] W). Values of corrected (465 [95] W) and uncorrected (444 [87] W) mean power were similar (p > 0.05). The mean bias value for all power parameters equated to less than +/-1% of the absolute values of power measured. Intraclass correlation coefficients for all data sets ranged from 0.97 to 0.98. Coefficients of variation for uncorrected and corrected values of peak power were 2.8 and 4.5%, while corresponding values for mean power were 2.9 and 3.2%, respectively. The reproducibility of all power indices was below 5%. The results of this study indicate that both uncorrected and corrected measurements of peak power output and mean power output can be used to assess performance during sprint arm ergometry.