Abstract: In order to balance the prediction accuracy of rockburst and engineering practicability, a prediction model of rockburst grade based on belief network was proposed. The mechanism of rockburst and the existing discrimination criterion were considered comprehensively, factors such as the maximum tangential stress and the ratio of uniaxial compressive strength of rock were chosen as the evaluation indexes. Based on the sample data set of expert argumentation, the input incomplete information was forecasted and optimized by the gradient descent method. By using the NETICA software, the existing sample information was statistically trained, and the probability distribution among nodes was obtained, the sensitivity of each criterion node to the target node was analyzed. Using the prediction model to analyze and forecast engineering project, which verifies the feasibility of the model providing a new method for rockburst grade prediction.