To investigate the applicability of different artificial intelligence(AI)flood foresting models in Chinas semi-arid and semi-humid regions, four types of AI models including decision tree(DT), multilayer perception(MLP), random forest(RF), and support vector machine(SVM)were selected to conduct hourly flood forecasting in three typical river basins of Shaanxi Province. Statistical metrics including the coefficient of correlation, Nash-Sutcliffe efficiency, root-mean-square error, mean absolute error and relative error are used to assess the models effectiveness in these typical basins for different forecasting periods. The results show that all models can achieve good performance in the semi-humid basins for short-term forecasting. However, the four AI models have relative lower accuracy in the semi-arid basins, and only the SVM model can achieve satisfactory forecasting accuracy. As the forecasting lead time increases, the performance of different models varies greatly. The SVM model is overall stable and has an obvious advantage for real-time flood forecasting in small and medium-sized basins. Performance of the RF and DT models declines slowly with increasing forecasting lead time, while performance of the MLP model decreases quickly with increasing lead time, showing lower stability.