In order to evaluate the stope stability scientifically and accurately, a two-stage fuzzy evaluation model of stope roof stability was established with the engineering practice of Sanshandao under-sea gold mining. Eleven indexes, i.e., compressive strength of rock mass, RQD, average joint spacing, groundwater condition, exposed area of roof, depth of stope, span of stope, span-depth ratio of stope, disturbance influence around, maximum exposure time of roof, support measures, were selected to be the evaluation indicators. Degree of membership of the continuous indicators was calculated using sectional-continuous functions and degree of the discrete indicators was obtained by fuzzy language variables. To avoid the influence of subjective factors, rough set theory was applied to determine the weight of indicators by training 12 sets of stope samples. Each of the 12 sets of stope samples was tested by the model, and the accurate ratio was 100%. Stability of another stope roof on the -400 m level was predicted by the model. The results show that the predicted stability classification was in good agreement with the actual situation. Therefore, the evaluation model can be used in the practical stope stability evaluation.