Simulation of local atmospheric flows around complex topography is important for several applications in wind energy (short-term wind forecasting and turbine siting and control), local weather prediction in mountainous regions and avalanche risk assessment. However, atmospheric simulation around steep mountain topography remains challenging, and a number of different approaches are used to represent such topography in numerical models. The immersed boundary method (IBM) is particularly well-suited for efficient and numerically stable simulation of flow around steep terrain. It uses a homogenous grid and permits a fast meshing of the topography. Here, we use the IBM in conjunction with a large-eddy simulation (LES) and test it against two unique datasets. In the first comparison, the LES is used to reproduce experimental results from a wind-tunnel study of a smooth three-dimensional hill. In the second comparison, we simulate the wind field around the Bolund Hill, Denmark, and make direct comparisons with field measurements. Both cases show good agreement between the simulation results and the experimental data, with the largest disagreement observed near the surface. The source of error is investigated by performing additional simulations with a variety of spatial resolutions and surface roughness properties.