The Jumahe River is one of the most important flood control areas in the Haihe River Basin， and it is difficult for flood forecasting. In this paper， the Xinanjiang Model， Modified Xinanjiang Model， rainfall flood model for Hebei Province and BP Neural Network Model were chosen to simulate the different historical floods occurred in the years of 1956， 1963 and 2012. The results show that the four kinds of flood models can be taken into use when the rain intensity is heavy and the rainfall duration is long； the relative errors of the peak flow simulation is less than 20%， and the accuracy of BP Neural Network Model is the best， all the models mentioned in this paper can be applied in arid and semi-arid regions.