The accuracy of smart meters for operation is closely related to the e vital interests of grid companies and the users，the numbers of the smart meters for operation is very large, about 550 million，covering 90% of the land area and the operation environment is very different. The real life of the smart meter is a big difference, so it is need to evaluate the real status of the smart meter and predict the rest of the life. On this paper, the smart meters for operation more than one year have been selected by stratified sampling method in Fujian Province and Heilongjiang Province, and a sampling life tolerance model is established. Considering the effect of temperature and humidity on error, the Weibull lifetime distribution model is established based on the bathtub curve. The proposed life prediction method was proved to be feasible through comparing with the test and prediction results in this paper.