Abstract: Five indices of the medium conditions, occurence background, and engineering factors of rocks, such as uniaxial compressive strength, rock quality distribution, coal mass strength, groundwater status, and monthly advance of working face are used in establishing, the dynamical evaluation model of the roof stability with artificial neural network(ANN). The subzone of roof stability is forecasted from data of working face advance of 40m, 60m, 80m and 100m per month, respectively. The result of ANN analysis of 5 indeces shows the factors effecting on the roof stability in a sequence from high to low degree are rock quality distribution, groundwater status, rock uniaxial compressive strength, coal mass strength and face advance. The measures for controlling different types of roof stability are suggested pertinently according to above results.