The sensitivity of ocean large-eddy simulations (LES) to model domain size and spatial resolution is systematically investigated in the context of a diurnal cycling oceanic boundary layer. The model domain size varies by a factor of 64, horizontal resolution varies by a factor of 12, and vertical resolution varies by a factor of 8. A control experiment with a high resolution serves as the truth by which other experiments with smaller domains and/or coarser resolutions are judged. During nighttime the primary balance in the turbulent kinetic energy budget is between buoyancy production, turbulent transport, pressure transport, and dissipation. During daytime the primary balance is between shear production, buoyancy destruction, and dissipation. It is found that mean fields and turbulent fluxes are insensitive to domain size as long as it is comparable to or greater than the mixed layer depth. The mean fields and turbulent fluxes are also insensitive to resolution, provided there are several vertical levels in the entrainment layer. Turbulent kinetic energy (TKE), however, shows varying degrees of sensitivity. Simulations with larger domains or with coarser resolutions tend to have larger values of TKE. Furthermore, TKE is more sensitive to domain size and resolution during the day (shear turbulence) than during the night (convection). Practically, it is better to use a larger domain with an anisotropic resolution than a small domain with an isotropic resolution, if computer resources are severely limited. With the same grid geometry, the former is not only cheaper to conduct but also can be more accurate as far as turbulent kinetic energy and turbulent fluxes are concerned. This underlines the importance of resolving the largest eddies in the oceanic boundary layer. We also conclude that validations of simple one-dimensional mixed layer models using LES solutions with a resolution of order 32¿32¿32 are not without merit in the sense that mean fields and turbulent fluxes are not much different when much higher resolution is used. However, if TKE is used as a variable of validation, higher resolution is needed. ¿ 2001 American Geophysical Union