The present report describes the development and validation of a turbulence model designed for atmospheric flows based on the concept of Large-Eddy Simulation (LES). The background for the work is the high Reynolds number k - #epsilon# model, which has been implemented on a finite-volume code of the incompressible Reynolds-averaged Navier-Stokes equations (RANS). The k - #epsilon# model is traditionally used for RANS computations, but is here developed to also enable LES. LES is able to provide detailed descriptions of a wide range of engineering flows at low Reynolds numbers. For atmospheric flows, however, the high Reynolds numbers and the rough surface of the earth provide difficulties normally not compatible with LES. Since these issues are most severe near the surface they are addressed by handling the near surface region with RANS and only use LES above this region. Using this method, the developed turbulence model is able to handle both engineering and atmospheric flows and can be run in both RANS or LES mode. For LES simulations a time-dependent wind field that accurately represents the turbulent structures of a wind environment must be prescribed at the computational inlet. A method is implemented where the turbulent wind field from a separate LES simulation can be used as inflow. To avoid numerical dissipation of turbulence special care is paid to the numerical method, e.g. the turbulence model is calibrated with the specific numerical scheme used. This is done by simulating decaying isotropic and homogeneous turbulence. Three atmospheric test cases are investigated in order to validate the behavior of the presented turbulence model. Simulation of the neutral atmospheric boundary layer, illustrates the turbulence model ability to generate and maintain the turbulent structures responsible for boundary layer transport processes. Velocity and turbulence profiles are in good agreement with measurements. Simulation of the flow over the Askervein hill is also performed. Speed-up and turbulence intensities show good agreement with measurements, except 400m downstream of the hill summit where speed-up is underestimated. Flow over a cube in a thick turbulent boundary layer is the final test case. The turbulence model ability to capture the physics of the large separated region downstream of the cube is demonstrated. The turbulence model is, however, shown to have trouble with very large values of roughness.