Abstract The sensitivity of high-resolution mesoscale simulations to boundary layer turbulence parameterizations is investigated using the fifth-generation Pennsylvania State University–NCAR Mesoscale Model (MM5) and observations from two field campaigns. Three widely used turbulence parameterizations were selected for evaluation, two of which [Blackadar (BK) and Medium Range Forecast (MRF) schemes] are simple first-order nonlocal schemes and one [Gayno–Seaman (GS) scheme] of which is a more complex 1.5-order local scheme that solves a prognostic equation for turbulence kinetic energy (TKE). The two datasets are the summer 1996 Boundary Layer Experiment (BLX96) in the southern Great Plains and the autumn 2000 Vertical Transport and Mixing (VTMX) field campaign in the Salt Lake Valley in Utah. Comparisons are made between observed and simulated mean variables and turbulence statistics. Despite the differences in their complexity, all three schemes show similar skill predicting near-surface and boundary layer mean temperature, humidity, and winds at both locations. The BK and MRF schemes produced daytime boundary layers that are more mixed than those produced by the GS scheme. The mixed-layer depths are generally overestimated by the MRF scheme, underestimated by the GS scheme, and well estimated by the BK scheme. All of the schemes predicted surface latent heat fluxes that agreed reasonably well with the observed values, but they substantially overestimated surface sensible heat fluxes because of a significant overprediction of net radiation. In addition, each parameterization overestimated the sensible and latent heat flux aloft. The results suggest that there is little gain in the overall accuracy of forecasts with increasing complexity of turbulence parameterizations.