A study of missing data reconstruction and predictability for precipitation in Beijing is completed in the period of 1841-1980. The new variable of precipitation with nonlinear transform has been done for following Gauss distribution. Further the method of stepwise regression with minimum of forecast error has been employed for reconstructing the missing data using the data of temperature and precipitation in earlier months and years as predictors. Three experimental schemes of using only precipitations or the composite of temperature and precipitation and sample insert value between data are adopted. The results show that the scherne of using the composite factors of temperature and precipitation has good predictability. The missing data in the period has been reconstructed.