A fog detection algorithm that uses geostationary satellite data has been developed and tested. This algorithm focuses on continuous fog detection since temporal discontinuities, especially at dawn and dusk, are a major problem with current fog detection algorithms that use satellite imagery data. This is because the spectral radiance at 3.7 µm contains overlapping emissive and reflectance components. In order to determine the radiance at 3.7 µm under fog conditions, radiative transfer model simulations were performed. The results showed that the radiance at 3.7 µm obviously varies with the solar zenith angle, and the brightness temperature differences between 3.7 µm and 10.8 µm are completely dissimilar between day and night (positive and varying with the angle during the daytime, but negative and constant at night). In this algorithm, a dynamic threshold is used as a function of the solar zenith angle. Moreover, additional criteria such as infrared, split-window channels, and a water vapor channel are used to remove high-level clouds. Also, the visible reflectance (0.67 µm) channel is used in the daytime algorithm because visible channel images are very practical for confirming a fog area with the high reflectivity and the smooth texture. The clear-sky visible reflectance for the previous 15 days was also employed to eliminate the surface effect that appeared during dawn and dusk. As the results, fog areas were estimated continuously, allowing the lifecycle of the fog system, from its development to decline, to appear obviously in the resulting images. Moreover, the estimated fog areas matched well with surface observations, except in a high latitude region that was covered by thin cirrus clouds.