In the process of mechanical assembly design, the essence of tolerance grade design is to correctly resolve the contradiction between the use requirements of mechanical parts and the manufacturing cost. The selection range of traditional tolerance grade is large, and most of them are manual look-up table selection and experience selection, so the uncertainty is large, and the accuracy and efficiency require the experience of the designer very high. In order to reduce the reliance on the designer’s experience in the selection of tolerance grades in tolerance matching, through comprehensive analysis of tolerance grade selection consideration factors, the fuzzy comprehensive evaluation principle was used to construct tolerance grade intelligent selection ontology rules, and a tolerance grade optimization inference model was established based on computer ontology language. The working process of the method was demonstrated through engineering examples, and the effectiveness of the method was verified.