In order to improve the validity of thunderstorm strength forecast, the data of lightning location are used to classify the thunderstorm strength from June to August, 2008 to 2010, and the radiosonde data are used to calculate 47 convective parameters to represent the characteristics of the Nanjing thunderstorm environment. The relationships of thunderstorm strength with 47 convective parameters are analyzed respectively. The convection parameters closely related to thunderstorm strength are selected as the forecasting predictors of thunderstorm strength. The Bayesian classification method and Logistic regression analysis are adopted to establish two thunderstorm forecasting models. By use of the testing samples, the forecasting models are tested and compared. The results indicate that the Hedike skill score of the Logistic regression analysis is 0396, can identify 30% of the severe thunderstorm, but that of the Bayesian classification’s is 037, can identify only 5% of the severe thunderstorms. It is obvious that Logistic regression has better indicative significance to thunderstorm potential strength forecast. The analysis of the nine convection parameters used to build the forecasting model, indicates that the stronger the thunderstorm activity, the warmer and moister the air at lower levels, the colder and drier the air at middle levels, the stronger the wind shear between lower and upper levels.