This study compared the accuracy and sensitivity of the Wet-Bulb Globe Temperature (WBGT) models developed based on domestic and international studies. The applicability of these models to formal weather observatory and user-installed sites such as the military training centers in Korea were evaluated. For the WBGT estimation, a physical model and four empirical models were used. A Liljegren model, which is known to have high accuracy, was used as the physical model. The KMA2006, JME (four-variable), KMA2016, and ABM (two-variable) models were used as the empirical models. The root-mean-square errors (RMSE) of ABM and Liljegren models obtained in this study are approximately 1 ℃ higher than RMSE results in US and New Zealand. The KMA2006 and JME models demonstrated high accuracy at RMSE of 0.45 ℃. The KMA2016 model performed better in the simulation than the physical model (RMSE of 1.06 ℃) even though only two variables were used. In particular, the KMA2016 model demonstrated good performance in the central region of the Korean peninsula, including the military training centers (RMSE of 0.75 ℃). However, it was underestimated in the southern part of the peninsula. Its performance became worse except for the time occurring maximum temperature. Therefore, at the Automated Surface Observing System stations where solar radiation and wind speed data are available, the KMA2006 and JME models can outperform other models. In the military, school, and industrial sites where only air temperature and relative humidity are observed, we expect that the WBGT can be estimated within an error range of ±0.6 ℃ using the KMA2016 model.