The accuracy of the conservative force-models is not clear and the new space accelerometers provide a unique opportunity for their validation. In this paper, the conservative-force model deficiencies are investigated via Principal Component Analysis (PCA) using 4 years of GRACE measurements (2006-2009). The deficiencies are assessed by comparing the accelerometer readouts with the differences between the conservative force-model and the precise orbit accelerations. Within this scheme, the spatiotemporal data analysis is synthesized in a time-series of grids whose latitudinal and longitudinal variations are respectively influenced by the half orbital period and the equatorial orbit shift. With maximum amplitude of 50 nm/s(2), the mean map shows a positive overestimation in Canada and Brazil and negative in Greenland. Explaining the 74% of the variability, the two first pairs of PCA modes show un-modeled patterns with amplitudes maxima of 80 nm/s(2) and 50 nm/s(2). Their periodic behavior and wave-length of degree-6 and degree-11s spherical harmonics strongly suggest an additional modeling and improvement. As for the long-term variations, our results show a latitudinal variation of 15 nm/s(2) maximum amplitude correlated with the day-night periods. Performing a combined analysis of ascending and descending orbits, a trend of 3 nm/s(2)yr in the Indian Ocean is also observed. Our new approach for orbital force models validation can be considered crucial for the current state-of-the-art of precise orbit determination (POD) and Time-Varying Gravity (TVG) modeling. (C) 2015 Elsevier Masson SAS. All rights reserved.