Abstract: Sea fog is a phenomenon of water vapor condensation or sublimation in marine atmospheric boundary layer and is also one of the main disastrous weathers on the coast of Guangdong Province in spring. However, there is no suitable method for operational sea fog forecasting in Guangdong due to the complexity of physical processes involved in the formation of sea fog. Therefore, historical sea fog reports from Shantou, Zhuhai and Zhanjiang surface meteorological observation and NCEP/NCAR FNL reanalysis for the period of 2000—2008 are analyzed to explore the feasibility of sea fog forecasting with a 24-hour lead time. The relationship between marine atmospheric conditions and sea fog events is examined by Classification and Regression Trees (CART), employing the NCEP/NCAR reanalysis data 24 hours before the sea fog events. Then, the decision tree models for sea fog forecasting are developed based on results of classification analysis. Finally, the physical significance of the forecasting rules is discussed based on existing theoretical knowledge on sea fog.The validation results by 10 cross-validation show that the forecasting accuracy of sea fog decision tree models developed by CART can reach 83.7%, 73.7% and 82.4% respectively for Shantou, Zhuhai and Zhanjiang on the coast of Guangdong Province. It can be interpreted or understood easily due to the clear logical relationship. The decision-making procedure can be developed and used directly to make fog/no-fog identification in operational sea fog forecasting with clear physical meanings. It also reflects the importance of the water vapor and the cooling effect of cold sea surface in the formation of advective cooling fog well. Simple calculation processes and relatively high classification accuracy make the CART an effective tool to develop sea fog forecasting model.