We quantify differences in ocean model simulations derived solely from atmospheric uncertainties and investigate how they relate to overall model errors as inferred from comparisons with data. For this purpose, we use a global configuration of the MITgcm to simulate 4 ocean solutions for 2000–2009 using 4 reanalysis products (JRA-25, MERRA, CFSR and ERA-Interim) as atmospheric forcing. The simulations are compared against observations and against each other for selected variables (temperature, sea-level, sea-ice, streamfunctions, meridional heat and freshwater transports). Forcing-induced differences are comparable in magnitude to model-observation misfits for most near-surface variables in the tropics and sub-tropics, but typically smaller at higher latitudes and polar regions. Forcing-derived differences are expectedly largest near the surface and mostly limited to the upper 1000m but can also be seen as deep as 4000m, especially in regions of deep water formation. Errors are not necessarily local in nature and can be advected to different basins. Results indicate that while forcing adjustments might suffice in optimization procedures of near-surface fields and at low-to-mid latitudes, other control parameters are likely needed elsewhere. Forcing-induced differences can be dominated by large spatial scales and specific time scales (e.g. annual), and thus appropriate error covariances in space and time need to be considered in optimization methodologies.