To solve the problem of low discrimination and low accuracy of reliability model selection of assumed test methods at present, one grey correlation model optimal selection method based on similar slope and closer median is put forward and the algorithm is studied. The reliability models of hydraulic system were established by 5 probability density functions based on the collected reliability data of hydraulic system of the computer numercial control (CNC) grinder. The proposed reliability model selection was carried out for fitting and optimal test of five different kinds of distribution models. The result showes that gamma distribution is most suitable for the hydraulic system. The discrimination is enhanced by 86.42%, 67.27% and 2.59% by the proposed model selection method compared by the correlation coefficient method, common grey correlation and Kolmogorov-Smirnov (K-S) method. The accuracy is very high and the model selection effect is good. The proposed method can be used to select the best reliability model in reliability design and it is also the reference to other model selection of curve fitting.