Abstract: On the basis of analysis of the factors influenci ng the stability of s oft rock roadway with different mechanical and physical features,a model to optimizing the support patterns of soft rock roadway and to pre dict its deformation is established by applying the theory of neural netwo rk. Its application in Meitian mining district shows that the model can select r ationally the support patterns of soft rock roadway and forecast accurately the deformation of sides and roof and floor of roadway, and that th e learning speed of network and the precision of model are enhanced with the app lication of reformatory BP algorithm.