Abstract: Existing roof stability assessment methods of coal roadway were summarized and analyzed, which included single index method and compound index method in classic methods and supervised learning method and unsupervised learning method in machine learning methods. It was pointed out that the classic methods assessed roof by single index or for a certain type of coal rock so that assessment result is incomplete or unreliable, and the machine learning method needed a large number of hand-crafted labeling of roof monitoring data with large workload and poor actual application effect. A new roof stability assessment mode of coal roadway based on generative adversarial network in deep learning was proposed according to advantage of extracting features from roof monitoring data automatically of the deep learningmethod, so as to decrease labor workload.