Abstract: The stability of a potential landslide involves the interaction and inter-coupling of multi-factors. So it is very difficult to evaluate the interaction relationship among the influencing factors quantitatively and determine weight considering multi-factors. Therefore according to other scholars' achievement and the principle of rock engineering system theory's matrix construction and decoding,a new potential landslide identification model is suggested based on improved BP network. Then,the arithmetic flow of coding interaction matrix and considering multi-factors determine weight are deduced. At last,the model is applied to identify potential soil landslides. The research results show that the parameters analysis of artificial neural network can be used not only to potential landslide identification but also to determine weight considering multi-factors.