Large-eddy simulation (LES) of atmospheric boundary layer (ABL) flow is performed over a homogeneous surface with different heat flux forcings. The goal is to test the performance of dynamic subgrid-scale models in a numerical framework and to compare the results with those obtained in a recent field experimental study (HATS (Kleissl et al., 2004)). In the dynamic model the Smagorinsky coefficient cs is obtained from test filtering and analysis of the resolved large scales during the simulation. In the scale-invariant dynamic model the coefficient is independent of filter scale, and the scale-dependent model does not require this assumption. Both approaches provide realistic results of mean vertical profiles in an unstable boundary layer. The advantages of the scale-dependent model become evident in the simulation of a stable boundary layer and in the velocity and temperature spectra of both stable and unstable cases. To compare numerical results with HATS data, a simulation of the evolution of the ABL during a diurnal cycle is performed. The numerical prediction of cs from the scale-invariant model is too small, whereas the coefficients obtained from the scale-dependent version of the model are consistent with results from HATS. LES of the ABL using the scale-dependent dynamic model give reliable results for mean profiles and spectra at stable, neutral, and unstable atmospheric stabilities. However, simulations under strongly stable conditions (horizontal filter size divided by Obukhov length >3.8) display instabilities due to basic flaws in the eddy viscosity closure, no matter how accurately the coefficient is determined.