Two approaches for the calibration of GRACE (Gravity Recovery And Climate Experiment) accelerometers are revisited. In the first approach, surface forces acting on the satellite are considered to derive the reference acceleration. In the second approach, the total acceleration consisting of a gravitational and a non-gravitational contribution is first determined from the reduced-dynamic orbits. The approximation of discrete satellite positions by a polynomial function allows the total acceleration to be obtained by a twofold derivative w.r.t. time. Calibration parameters (scale factor and bias) and statistical values are estimated for periods with a low and high solar activity. The quality of these two approaches shows dependencies on solar activity and consequent variations in the magnitude of the non-gravitational reference acceleration. Besides, the quality of the presented results is affected by the orientation of the orbital plane w.r.t. the Sun. The second approach is vitiated by a periodic disturbing signal on cross-track axis. This signal has been pointed out in earlier studies (Calabia et al., Aerosp Sci Technol 45, 2015; Calabia and Jin, Aerosp Sci Technol 49, 2016). We apply a moving window median filter to recover the underlying non-gravitational signal for accelerometer calibration. The calibration is accomplished by a direct comparison of reference accelerations and observed accelerometer measurements without introducing any a priori values or constraints. The focus of this work is more sensor oriented than gravity field recovery (GFR) related. Nevertheless, the results can be used as initial values for precise orbit determination (POD) or for pre-processing of accelerometer measurements in a multi step gravity field recovery approach (Klinger and Mayer-Gürr, Adv Space Res 58(9), 2016). The final paper is available at https://doi.org/10.1007/1345_2018_46.