A forecasting expert system for heavy rain in Guangdog province during its first flood season is introduced in this paper. The system, which develops on IBM PC/XT computer with dBASE Ⅲ and compiler BASIC language, is based on aser-ies of fairly useful experiences concerning torrential rain forecasts for the season. Supported by its data base, rule base and related service programs, the system is capable of simulating forecaster's way of thinking to perform the yes/no judgement and forecasting where it would have heavy rain by simple reasoning. Driving a multilevel menu, the forecaster can arbitrarily and conveniently change various parameters for obtaining intermidiate results and/or some other valuable charts and tables as well, before the man-machine interactional forecasting conclusion is reached. Besides, a forced-study sub-system is set up to provide the system with a capability of renewing experiential rules. A two-year operation verification since April 1,1986 proved that the successrate of yes/no and falling area forecasts of the expert system are 68.8% and 66.6%, respectively.