The aim of this study was to compare equivalence and agreement of physical activity output data collected by a Research Tri-axial accelerometer (R3T) during walking and running on a treadmill versus on land. Fifty healthy volunteers, 35 males (age 21.9 +/- 1.8 years) and 15 females (age 21.6 +/- 0.7 years), underwent a series of tests on a treadmill and on land with the order of testing administered randomly. Each participant walked for 10 min at 4 km x h(-1) and 6 km x h(-1), and ran at 8 km x h(-1) and 10 km x h(-1), with the same accelerometer. Analysis of output data was assessed by two statistical tests: the equivalence test and Bland and Altman method. Mean differences for walking were 41.2 +/- 129.8 counts per minute and -68.8 +/- 173.15 counts per minute at 4 km . h(-1) and km x h(-1), respectively. Mean differences for running were 19.1 +/- 253.20 counts per minute and 38.9 +/- 270.2 counts per minute at 8 km x h(-1) and 10 km x h(-1), respectively. The physical activity output data from the treadmill were higher by an average of 3.5% than the data collected on land. The differences obtained between the treadmill and on land were small and non-significant. The equivalence test showed that output data from the treadmill versus on land were equivalent (P < 0.05). The Bland and Altman method showed good agreement between the counts obtained on the treadmill and on land (P < 0.05). In conclusion, physical activity output data were similar as measured by the RT3 accelerometer on a treadmill and on land. The findings suggest that the RT3 may be used in a laboratory and extrapolated to data obtained on land.