The emergence of high resolution satellite measurements of the gravitational field (GOCEmission) offers promising perspectives for the study of the Earth’s interior. These new datacall for the development of innovant analysis and interpretation methods. Here we combine a forward prism computation with a Bayesian resolution approach to invert for these gravity gradient data configuration. We apply and test our new method on satellite data configuration, i.e. 225 km height with a global and homogeneous geographic distribution. We first quantify the resolution of our method according to both data and parameterization characteristics. It appears that for reasonable density contrast values (0.1 g.cm−3) crustal structures have to be wider than ∼28 km to be detectable in the GOCE signal. Deeper bodies are distinguishable for greater size (35 km size at 50 km depth, ∼80 km at 300 km depth). We invert the six tensor components, among which five are independent. By carefully testing each of them and their different combinations, we enlighten at rade off between the recovery of data and the sensitivity to inversion parameters. We particularly discussed this characteristic in terms of geometry of the synthetic model tested (structures orientation, 3-D geometry, etc.). In terms of RMS value, each component is always better explained if inverted solely, but the result is strongly affected by the inversion parameterization (smoothing, variances, etc.). On the contrary, the simultaneous inversion of several components displays a significant improvement for the global tensor recovery, more dependent on data than on density variance or on smoothness control. Comparing gravity and gradient inversions, we highlight the superiority of the GG data to better reproduce the structures especially in terms of vertical location. We succesfully test our method on a realistic case of a complex subduction case for both gradient and gravity data. While the imaging of small crustal structures requires terrestrial gravity dataset, the longest wave length of the slab is well recovered with both data sets. The precision and homogeneous coverage of GOCE data however, counterbalance the heterogeneous and often quite non existance coverage of terrestrial gravity data. This is particularly true in large areas which requires a coherent assemblage of heterogeneous data sets, or in highr elief, vegetally covered and offshore zones