The non-motorized treadmill system initially reported by Lakomy in 1984 has been used extensively to assess sprinting performance. However, there has been limited research into the reliability of power output measurement using such systems. The aim of this study was to design a system and protocol capable of measuring treadmill sprinting performance in rugby players and to assess the reliability of this system for measuring power output. Twenty-seven rugby players, all of whom were familiar with treadmill sprinting, performed three maximal 6 s sprints with 2 min recovery between sprints, on two occasions 1 week apart. Both tests were performed on a non-motorized Woodway tramp treadmill, interfaced to a data acquisition system. There were no significant differences (P > 0.05) between power output for repeated trials on the same day (between trials) or for repeated trials on different days (between days). Limits of agreement for maximum average power (the average of 100 readings per second) were 4+/-98 and 30+/-157 W for between trials and between days, respectively. When reported as ratio limits of agreement, these were 1.07 (*/divided 1.12) and 1.03 (*/divided 1.16), respectively. The limits of agreement for maximum instantaneous power (the highest of 100 readings per second) were 51+/-464 and 105+/-588 W for between trials and between days, respectively. When reported as ratio limits of agreement, these were 1.02 (*/divided 1.20) and 1.04 (*/divided 1.21) for between trials and between days, respectively. The coefficients of variation for all measures of power output were less than 9.3%. Hence, the treadmill system and protocol developed in this study provide a reliable measure of power output for rugby players.