For better modeling the variations in the vertical distribution of water vapor, in this study, a new function for the vertical variation in water vapor was derived, named lapseRPWV. From the analyses of lapseRPWV time-series, it was found that its vertical distribution is strongly correlated with the relative magnitude of total precipitable water vapor (TPWV). This study proposed a method that used six data ranges of TPWV to determine the relative magnitude of TPWV. For the periodic variations in the classified lapseRPWV time-series in each of six TPWV ranges, a spatio–temporal lapseRPWV model was developed for each range. The new models were validated by comparing their predictions against the references from sounding data at 12 radiosonde stations in China, and their performances were also compared with that of the commonly used water vapor scale height (H) model. Results showed that, first, the number of stations that had reduced annual RMSE of H values in TPWV ranges from 1 to 6 accounted for 92%, 92%, 67%, 83%, 100%, and 100% of the total stations, respectively. Second, the proportions of the height range that had reduced annual RMSE of water vapor density (WVD) in all height ranges within all TPWV ranges were above 75% at the 12 stations. Last, considering all TPWV ranges as a whole, in each of 10 height ranges, the annual RMSEs of WVD of all the stations reduced at least 11%, 20%, 43%, 48%, 40%, 38%, 32%, 35%, 32%, and 28%, respectively.