By means of the classical Lee-Carter mortality model and Chinese mortality data from 1994 to 2019, this paper constructs the Lee-Carter mortality model to study the prediction of Chinese life expectancy at birth. The research shows that although Chinese mortality data sources are get rich step by step, the gap and conflict of data quality from different sources are still obvious. With the decline of mortality level, the quality problem of mortality data in 1‰ population change sampling survey since 1994 is more prominent. From the fitting effect of Lee-Carter mortality model, the age group with small average relative error in fitting age-specific mortality is 40-84 years old, and the fitting effect of 0-year-old group is poor. The data quality of 0 - year-old population mortality is very important in the process of model fitting, calculation and prediction of life expectancy at birth. Due to the error offset of age-specific mortality data, the estimation error of life expectancy at birth decreased significantly. In the application of the research results of life expectancy at birth, the relative error and absolute error of life expectancy at birth of population for different ages need to be treated differently. Considering the quality of Chinese historical data, it is necessary to prevent the model from over fitting in the modeling process. Since there is little possibility of re-reporting, especially over reporting, the existing data and models must overestimate or over prediction the life expectancy at birth.
