Abstract It is desirable to find a numerical forecast model that can better simulate and forecast the evolution of radiation fog. Here, observations made on 21 December 2015 at Nanjing University of Information Science and Technology (NUIST) are used to validate simulations from the Weather Research and Forecast (WRF) model and WRF nested Large-Eddy Simulation (WRF-LES) model. The WRF-LES model is significantly better than the WRF model at simulating various meteorological factors at the near-surface. Especially for the oscillation characteristics of meteorological factors, the WRF-LES model provides a more detailed and accurate description. The differences in the temporal evolution of visibility in the two models are evaluated during the radiation fog process. Compared with the WRF model, the fog onset time and the start of the strong phase predicted by the WRF-LES model are both more consistent with the actual situation. The temperature inversion (TI) layer from the WRF-LES model is lower than that of the WRF model, but its strength is closer to the observations. In addition, the WRF-LES model can better demonstrate the evolution of the planetary boundary layer height (PBLH) over time in the life cycle of this radiation fog than the WRF model. Based on the simulation results of the TI structure and the PBLH, it is shown that improving the ability of the numerical model to predict the turbulent mixing intensity can improve its ability to predict radiation fog. Meteorological station data are combined with experimental data to evaluate the ability of the two models to simulate a horizontal fog area. Combining the four indices with the horizontal distribution of the simulated fog region, the WRF-LES model is shown to be more capable of predicting radiation fog than the WRF model.