The present study examined the reliability of work and power measures during a 5 × 6-s cycle ergometer test of repeated-sprint ability. Nine, well-trained, female soccer players performed five, 5 × 6-s repeated-sprint tests on a front-access cycle ergometer on separate days. Sprints were separated by 24 s of active recovery. Absolute measures of total work done (W tot), total peak power (PPtot), work done during sprint 1 (W 1) and peak power output during sprint 1 (PP1) were recorded. Decrement scores in work done (W dec) and peak power output (PPdec), and fatigue indices for work done (FI W ) and peak power (FI P ), were calculated. Significant improvements in all of the work and power measures were observed between trial 1 and subsequent trials (P < 0.05), but no significant differences were identified between trials 2, 3, 4 and 5. The same was true for increases in the decrement scores. The coefficient of variation (CV) was established to reflect within-subject reproducibility for each variable. The CV was significantly improved by the third trial for work done (W tot CV: trials 1–2 = 5.5%; trials 3–4 = 2.8%), peak power (PPtot CV: trials 1–2 = 5.1%; trials 3–4 = 2.7%) and performance decrement scores (P < 0.05). The standard error of measurement (SEM) and intraclass correlation coefficient (ICC) were also calculated for each variable and expressed within 95% confidence intervals. It was concluded that two familiarisation trials are optimal for collecting reliable data from a 5 × 6-s repeated-sprint cycling test. Furthermore, due to the large variation around performance decrement it was suggested that decrement scores ought to be interpreted with caution.