We provide global fields of simulated ocean velocity in zonal (netCDF variable UKO) and meridional (netCDF variable VKE) direction at a depth of 420 m. Six climate states are covered in the data set:1. data set PI_mpiom_UKO_VKE_timmean_420m.nc: pre-industrial (PI) control state (representative for 1850 AD) as used in the publications by Stepanek and Lohmann (2012) and Zhang et al. (2013). The respective data is courtesy of Zhang et al. (2013)2. data set LGM_mpiom_UKO_VKE_timmean_420m.nc: a climate state of the Last Glacial Maximum (simulation LGM-W by Zhang et al., 2013), representative for 21 kiloyears (ka) before present (BP)3. data set Plio_mpiom_UKO_VKE_timmean_420m.nc: a climate state of the Mid-Pliocene Warm Period (simulation experiment 2 by Stepanek and Lohmann, 2012), that covers the time from 3.29 - 2.97 million years (Ma) BP4. data set MIO_mpiom_UKO_VKE_timmean_420m.nc: a climate state representing conditions of the early to middle Miocene (23 to 15 Ma BP) including a regional bathymetry reconstruction (15 Ma) of the North Atlantic / Arctic Ocean by Ehlers and Jokat (2013) and considering 450 ppm of carbon dioxide (simulation EO450 by Stärz et al., 2017)5. data set MioW_mpiom_UKO_VKE_timmean_420m.nc: a climate state representing conditions of the early to middle Miocene (23 to 15 Ma BP) including a regional bathymetry reconstruction (15 Ma) of the Weddell Sea (Huang et al., 2014) and considering 450 ppm of carbon dioxide in the atmosphere (simulation MIOW_450 by Huang et al., 2017)6. data set MioW_PIS_mpiom_UKO_VKE_timmean_420m.nc: a climate state similar to 5. but with PI (278 ppm) carbon dioxide concentration and prescribed modern ice sheets (simulation MIOW_PIS by Huang et al., 2017)All data sets represent climatological annual averages over a time period of 100 years. The oceanography is based on climate simulations performed with the Community Earth System Models (COSMOS) that consist of the atmosphere general circulation model ECHAM5 (Roeckner et al., 2003), internally coupled to the land surface and terrestrial carbon cycle model JSBACH (Raddatz et al., 2007) in T31 resolution (3.75°x3.75°) with 19 vertical levels on a hybrid sigma-pressure coordinate, and the ocean general circulation model MPIOM (Marsland et al., 2003) on a bipolar curvilinear GR30 grid with a formal resolution of 3.0°x1.8° and 40 z-coordinate levels. Exchange of momentum, mass, and energy between the atmosphere and ocean domain is enabled via the OASIS3 coupler (Valcke et al., 2003).For all simulations, ocean characteristics (including sea ice) and properties of atmosphere and land (including the land carbon cycle and dynamic vegetation) are computed based on the prescribed climate forcing (concentration of atmospheric trace gases, configuration of the Earth's orbit) and boundary conditions (land surface elevation, ice sheets, ocean bathymetry, and land sea mask). For details of the utilized boundary conditions and climate forcing refer to the original publications describing the data (Stepanek and Lohmann, 2012; Zhang et al., 2013; Huang et al., 2017; Stärz et al., 2017).In case of analyzing velocities of the Arctic Ocean, note that for few grid cells in the northernmost data row (89.5°N), between 280°E and 293°E, there is a data artifact in variable VKE. This artifact is a side effect of rotating and interpolatiing velocities from the curvilinear model grid to a standard NSWE coordinate system; this artifact has not been removed from the data sets.