Through the statistical analysis of different types of severe convective weather from April to September from 1998 to 2009 in Shanghai, a stepwise regression method is used to develop 0 to 12 hour potential forecast equations for all types of severe convective weather by using 42 convection parameters and their time variation. A new approach to convective potential forecasting based on the classification of four key convective parameters is designed. The key convective parameters are K index, SI, PWV (Precipitable Water Vapor) and θsedif85 (difference between 500 and 850 hPa in pseudo equivalent temperature), which depict the atmospheric thermal and water vapor conditions, respectively. According to the distributions of different types of convection, these four convective parameters are classified into 3 levels, and convective potential forecast equations are developed for each level, respectively. In comparison with the original equations, the forecast equations based on classified convective parameters are with dramatic increasing validity in the forecasting of thunderstorms, high winds, severe thunders and all types of convective weather. In addition, better performance would be granted with the following combinations: the classified SIbased equations for forecasting thunderstorms, classified PWVbased equations for forecasting severe thunder cases and all types of convective weather. The optimal combination method of severe convective potential forecasting based on the classification of key convection parameters has been used in routine application of numerical weather prediction model outputs.