In order to reduce flood loss and improve the prediction accuracy of water inrush from coal seam floor, a prediction model of water inrush grade in Feicheng mining area was established based on the mine hydrogeological data of this area, selecting six indexes including water pressure, development degree of floor fissure, fault gap, karst development degree, aquifer thickness and aquiclude thickness. The method of mean impact value (MIV) was used to evaluate the influence of each variable in the neural network model on the prediction of water inrush grade. The results show that this model has high prediction accuracy. Small water inrush in Feicheng mining area is mainly determined by the attributes of the aquifer and aquiclude. The occurrence of large and extra-large water inrush is closely related to the factors such fault structure, karst and development degree of floor fissure.