The present study compared the accuracy of triaxial accelerometry and the doubly labelled water (DLW) method for measuring physical activity (PA) in Japanese adolescents. A total of sixty adolescents aged 12–15 years were analysed. The total energy expenditure (TEE) was measured over 7 d by the DLW method and with an EW4800P triaxial accelerometer (Panasonic Corporation). The measured (RMRm) and predicted RMR (RMRp) were 5·7 (sd 0·9) and 6·0 (sd 1·0) MJ/d, respectively. TEE measured by the DLW method and accelerometry using RMRm or RMRp were 11·0 (sd 2·6), 10·3 (sd 1·9), and 10·7 (sd 2·1) MJ/d, respectively. The PA levels (PAL) measured by the DLW method using RMRm or RMRp were 1·97 (sd 0·31) and 1·94 (sd 0·31) in subjects who exercised, and 1·85 (sd 0·27) and 1·74 (sd 0·29) in subjects who did not exercise. The percentage of body fat correlated significantly with the percentage difference between RMRmv. RMRp, TEE, PA energy expenditure (PAEE) and PAL using RMRp, and PAL using RMRm assessed by the DLW method and accelerometry. The present data showed that while accelerometry estimated TEE accurately, it did not provide the precise measurement of PAEE and PAL. The error in accelerometry was attributed to the prediction error of RMR and assessment in exercise.