The weighted mean tropospheric temperature is a critical parameter in the conversion of wet zenith delay to precipitable water vapor in GPS Meteorology. This parameter can not be calculated from the radiosonde data in real time through the conventional methods. In this study, we first discuss the admissible error of weighted mean temperature to enable the accuracy of the conversion better than 1 mm, then summarize the performance of some of the existing methods. An empirical formula is established that satisfies the real-time requirement in GPS meteorology using Sequential Regression Analysis method. It is shown that this real-time formula as compared with other empirical methods is more accurate for local applications.