An aerospace software defect prediction model PRM_METHOD was proposed by use of the advantage of probabilistic relational model in describing and reasoning the relationship between multi-attribute classes and their uncertainty knowledge. First, a software defect classification method based on software test was proposed, and the theoretical basis of the application of probabilistic relational model to the aerospace software defect prediction was analyzed via the relationship between software defect classes. Then, under the definition and generalization of staff capacity and the feature of defect quantity, the model PRM_METHOD was described with its structure, learning and predict process. Moreover, an improved k-average clustering algorithm based on closing data gap was proposed aim at data set classification operation. Finally, an aerospace software was taken as the example to actualize the model, and the practical testing data were used as the training set and validation set to validate it as well. The results show that the average of mean absolute deviation between the validation set and predict result is 0.086 8, which means the prediction accuracy of the model is 0.913 2. Therefore, the conclusion is that the model PRM_METHOD has better prediction accuracy to the aerospace software defect prediction with a more complex associated relationship.