Aiming at the problem of the lower discrimination of fit goodness test in the reliability modeling of CNC grinder, a reliability model optimization method based on the minimum distance between sample data and fitting curve was proposed. Based on the collected fault data, five different distribution functions were used to model the reliability of the hydraulic system of CNC grinder, and the fit goodness of each distribution model was tested by using the existing fit goodness test methods and the proposed minimum distance method respectively. The results show that the optimal reliability model of the hydraulic system of CNC grinder is gamma distribution; compared with grey correlation method and K-S test, the minimum distance method not only has accurate optimization result, but also improves the average discrimination by 99.6% and 135.6% respectively.