SUMMARY:We analyse a hierarchy of three 1/4° global numerical simulations (ORCA-025.G70 (G70), ORCA-025. G85 (G85) and GLORYS1V1 (GLORYS)) by assessing their performance against observational data in the western Mediterranean. When compared with the EN3_v2a temperature and salinity database, the simulations are capable of reproducing surface layer temperature interannual variability but G70 is inaccurate with intermediate and deep-layer trends. This aspect is improved by the increased vertical resolution of G85 and by data assimilation in GLORYS. Salinity is the most problematic parameter because of the imbalance of the freshwater budget derived from inaccuracies in the atmospheric forcing parameters. Surface salinity restoring is needed in order to avoid salinity drift and inaccurate sea-level trends. G70, with a stronger relaxation, has a lower trend closer to altimetric measurements than G85. Mean surface circulation is well reproduced for relatively large-scale signals. We further show that G85 and GLORYS provide evidence of the 2004-2005 and 2005-2006 deep convection events in the Gulf of Lion. Finally, transports through the main straits of the western Mediterranean are correct in order of magnitude, direction and seasonal cycle when compared with observations. This study contributes to the improvement of the ORCA hierarchy of simulations and points out the strengths and weaknesses of these simulations in the Mediterranean Sea.