In this study, fog simulations were conducted using the Fifth-Generation NCAR/Penn State Mesoscale Model (MM5) in and around the Yodo River Basin, Japan. The purpose is to investigate the MM5 performance of fog simulation for long-term periods. The simulations were performed for January, February, March, and July, 2005 with a coarse 3-km and a nested fine 1-km grid domains. Results of the simulations were compared with data from ten meteorological observatories, fog sampling site in Mt. Rokko, and visibility measurement sites along the Second Meishin Expressway. At the meteorological observatories, the MM5 predictions agreed well with the observed temperature and specific humidity. In the Mt. Rokko region, MM5 generally reproduced the occurrence of relatively thick fog events but tended to overestimate liquid water content (LWC) of fog (by factors of 2.2-3.3 in terms of monthly mean LWC). In the Second Meishin Expressway region, while MM5 identified the specific sites at which fog either frequently or seldom occurs, the model underestimated the monthly fog frequencies by factors of more than 1.5. Overall, MM5 reproduced the general trend of fog events, and the model performance may be improved by using more adequate land surface data and suitable physics options for our study.