To evaluate whether an activity monitor based on body acceleration measurement can accurately assess the energy cost of the human locomotion, 12 subjects walked a combination of three different speeds (preferred speed +/- 1 km/h) and seven slopes (-15 to +15% by steps of 5%) on a treadmill. Body accelerations were recorded using a triaxial accelerometer attached to the low back. The mean of the integral of the vector magnitude (norm) of the accelerations (mIAN) was calculated. VO2 was measured using continuous indirect calorimetry. When the results were separately analysed for each incline, mIAN was correlated to VO2 (average r = 0.87, p<0.001, n = 36). VO2 was not significantly correlated to mIAN when data were globally analysed (n = 252). Large relative errors occurred when predicted VO2 (estimated from data of level walking) was compared with measured VO2 for different inclines (-53% at +15% incline, to +55% at -15% incline). It is concluded that without an external measurement of the slope, the standard method of analysis of body accelerations cannot accurately predict the energy cost of uphill or downhill walking.