We develop, calibrate and test a dataset intended to drive global ocean hindcasts simulations of the last five decades. This dataset provides surface meteorological variables needed to estimate air-sea fluxes and is built from 6-hourly surface atmospheric state variables of ERA40. We first compare the raw fields of ERA40 to the CORE.v1 dataset of Large and Yeager (2004), used here as a reference, and discuss our choice to use daily radiative fluxes and monthly precipitation products extracted from satellite data rather than their ERA40 counterparts. Both datasets lead to excessively high global imbalances of heat and freshwater fluxes when tested with a prescribed climatological sea surface temperature. After identifying unrealistic time discontinuities (induced by changes in the nature of assimilated observations) and obvious global and regional biases in ERA40 fields (by comparison to high quality observations), we propose a set of corrections. Tropical surface air humidity is decreased from 1979 onward, representation of Arctic surface air temperature is improved using recent observations and the wind is globally increased. These corrections lead to a significant decrease of the excessive positive global imbalance of heat. Radiation and precipitation fields are then submitted to a small adjustment (in zonal mean) that yields a near-zero global imbalance of heat and freshwater. A set of 47-year-long simulations is carried out with the coarse-resolution (2° × 2°) version of the NEMO OGCM to assess the sensitivity of the model to the proposed corrections. Model results show that each of the proposed correction contributes to improve the representation of central features of the global ocean circulation.