With the gradual improvement of the reliability level for the system, the sample size of failures is different, and the MTBF (Mean Time Between Failure) of different systems also has great differences. In view of the difference in accuracy when using different reliability modeling methods to evaluate the reliability of different MTBF' systems, this paper proposes use MTBF as an index to select reliability modeling and evaluation methods. In an example, the effectiveness of the method proposed in this paper is verified by modeling and analysis of the Weibull distribution and the power-law distribution.