Common non-invasive measurements of muscle fatigue during dynamic contractions are based on the surface electromyography (SEMG). However, these measurements include movement artefacts such as the change of muscle length and the movement of electrodes. These artefacts significantly contribute to the non-stationary properties of the measured SEMG signal. Consequently, the measurements of the muscle fatigue employ various sensors and signal-processing methods. In this paper, we present a method for the estimation of the fatigue which uses an accelerometer to determine the SEMG segments within which the movement artefacts are resolved and the wide-sense stationarity holds. The fatigue is then estimated by the median frequencies of their power spectra. To illustrate the features of the proposed method, we apply it to the biceps curl exercise.