Geoid models for the Mediterranean were computed using the remove-compute-restore method and Stokes-FFT, using shipborne gravity or altimetry inferred gravity data over sea and land gravity data. The remove step over sea does not include residual terrain correction (bathymetry), which leads to slightly worse results. The models were compared to an independent geoid constructed by subtracting the Mean Dynamic Topography from the Mean Sea Surface, and secondly to drifter-observed current speeds. Results revealed significant errors in the gravimetric geoid at smallest scales, and analysis of the results of this intermediate model showed that improvement is required in the gravity data preprocessing, specifically the de-biasing of marine data, as well as the gridding (interpolation) procedure. These issues will be addressed before the release of the final geoid model early 2018. Based on the drifter comparisons, the geoid based on altimeter data is the most accurate, more accurate than EIGEN6C4, and notably so at scales less than 50 km.