We compare the total kinetic energy (TKE) in four global eddying ocean circulation simulations with a global dataset of over 5000, quality controlled, moored current meter records. At individual mooring sites, there was considerable scatter between models and observations that was greater than estimated statistical uncertainty. Averaging over all current meter records in various depth ranges, all four models had mean TKE within a factor of two of observations above 3500. m, and within a factor of three below 3500. m. With the exception of observations between 20 and 100. m, the models tended to straddle the observations. However, individual models had clear biases. The free running (no data assimilation) model biases were largest below 2000. m. Idealized simulations revealed that the parameterized bottom boundary layer tidal currents were not likely the source of the problem, but that reducing quadratic bottom drag coefficient may improve the fit with deep observations. Data assimilation clearly improved the model-observation comparison, especially below 2000. m, despite assimilated data existing mostly above this depth and only south of 47°N. Different diagnostics revealed different aspects of the comparison, though in general the models appeared to be in an eddying-regime with TKE that compared reasonably well with observations. © 2010 Elsevier Ltd.