This paper assesses the performance of the Weather Research and Forecasting Model (WRF) as a tool for multiscale atmospheric simulations. Tests are performed in real and idealized cases with multiple configurations and with resolutions ranging from the mesoscale (gridcell size ~10 km) for the real cases to local scales (gridcell size ~50 m) for both real and idealized cases. All idealized simulations and the finest real-case simulations use the turbulence-resolving large-eddy simulation mode of WRF (WRF-LES). Tests in neutral conditions and with idealized forcing are first performed to assess the model’s sensitivity to grid resolutions and subgrid-scale parameterizations and to optimize the setup of the real cases. An increase in horizontal model resolution is found to be more beneficial than an increase in vertical resolution. WRF-LES is then tested, using extensive observational data, in real-world cases over complex terrain through nested simulations in which the mesoscale domains drive the LES domains. Analysis of the mesoscale simulations indicates that the data needed to force the largest simulated domain and to initialize surface parameters have the strongest influence on the results. Similarly, LES model fields are primarily influenced by their mesoscale meteorological forcing. As a result, the nesting of LES models down to a 50-m resolution does not improve all aspects of hydrometeorological predictions. Advantages of using fine-resolution LES are noted at nighttime (under stable conditions) and over heterogeneous surfaces when local properties are required or when resolving small-scale surface features is desirable.