It is presented to predict slope splash-erosion for rain drops based on rough neural network. The redundancy information is reducted by the relative dependability between condition attribute and decision attribute of rough sets. The test indexes such as water depth and flux are deleted. The 2−5−1 rough neural network is established in which input parameters are gradient and rain intensity while output parameter is splash-erosion quantity. The frame of neural network is predigested. The train time of neural network is decreased. The reduced sloping field rain drops splash-erosion linear regression correlation coefficient of prediction and experiment is larger than that had not been reduced. The constringency speed is faster than that had not been reduced. The example calculation indicates that the rough neural network is an efficient and feasible algorithm to forecast sloping field rain drops splash-erosion.