The purpose of this study was to establish the accuracy of five published accelerometer regression equations that predict time spent in different intensity classifications during free-living activities. Ten participants completed physical tasks in a field setting for a near-continuous 5 - 6 h-period while oxygen uptake and accelerometer data were collected. The amount of time spent in resting/light, moderate and hard activity was computed from 3 and 6 MET cut-points associated with five existing regression formulas relating accelerometer counts x min -1 to energy expenditure. The Freedson cut-points over-estimated resting/light activity by 34 min (13 %) and under-estimated moderate activity by 38 min (60 %). The Hendelman cut-points for all activities underestimated resting/light activity by 77 min (29 %), and overestimated moderate activity by 77 min (120 %). The Hendelman cut-points developed from walking activities over-estimated resting/light activity by 37 min (14 %) and under-estimated moderate activity by 38 min (60 %). Estimates from the Swartz cut-points for estimating time spent in resting/light, moderate and hard intensity activity were not different from the criterion measure. The Nichols cut-points over-estimated resting/light activity by 31 min (12 %) and under-estimated moderate activity by 35 min (55 %). Even though the Swartz method did not differ from measured time spent in moderate activity on a group basis, on an individual basis, large errors were seen. This was true for all regression formulas. These errors highlight some of the limitations to using hip-mounted accelerometers to reflect physical activity patterns. The finding that different accelerometer cut-points gave substantially different estimates of time spent data has important implications for researchers using accelerometers to predict time spent in different intensity categories.