Purpose The aim of this study was to test the validity of a method using an inertial measurement unit for estimating activity-related energy expenditure (AEE) during walking in middle-aged adults. Methods Twenty healthy middle-aged participants completed different treadmill walking trials with an inertial measurement unit adhered to their lower back. Gas exchange was monitored with indirect calorimetry. Mechanical data were used to estimate AEE from an algorithm developed by Bouten et al. (Med Sci Sport Exer 26(12):1516–1523, 1994). Three methods for removing the gravitational component were proposed and tested: mean subtraction method (MSM), high-pass filter method (HPM) and free acceleration method (FAM). Results The three methods did not differ significantly from the indirect calorimetry [bias = − 0.08 kcal min−1; p = 0.47 (MSM), bias = − 0.08 kcal min−1; p = 0.48 (HPM) and bias = − 0.15 kcal min−1; p = 0.23 (FAM)]. Mean root mean square errors were 0.43, 0.42 and 0.51 kcal min−1 for MSM, HPM and FAM, respectively. Conclusion This study proposed an accurate method for estimating AEE in middle-aged adults for a large range of walking intensities, from slow to brisk walking, based on Bouten’s algorithm.