Abstract The performance of mesoscale meteorological models is evaluated for the coastal zone and Los Angeles area of Southern California, and for the San Joaquin Valley. Several configurations of the Weather Research and Forecasting Model (WRF) with differing grid spacing, initialization, planetary boundary layer (PBL) physics, and land surface models are compared. One configuration of the Coupled Ocean–Atmosphere Mesoscale Prediction System (COAMPS) model is also included, providing results from an independent development and process flow. Specific phenomena of interest for air quality studies are examined. All model configurations are biased toward higher wind speeds than observed. The diurnal cycle of wind direction and speed (land–sea-breeze cycle) as modeled and observed by a wind profiler at Los Angeles International Airport is examined. Each of the models shows different flaws in the cycle. Soundings from San Nicolas Island, a case study involving the Research Vessel (R/V) Atlantis and the NOAA P3 aircraft, and satellite images are used to evaluate simulation performance for cloudy boundary layers. In a case study, the boundary layer structure over the water is poorly simulated by all of the WRF configurations except one with the total energy–mass flux boundary layer scheme and ECMWF reanalysis. The original WRF configuration had a substantial bias toward low PBL heights in the San Joaquin Valley, which are improved in the final configuration. WRF runs with 12-km grids have larger errors in wind speed and direction than those present in the 4-km grid runs.