Abstract: The Regional Meso-scale Numerical Prediction System of Northeast China (RMNPSNC) is used to study the application of an observation error diagnostic method based on Desroziers et al (2005) (here in after Des method) in the GNSS-ZTD (Global Navigation Satellite System-Zenith Total Delay, ZTD) 3DVar assimilation. The Des method is compared with traditional observation error determination method based on ZTD assimilation and forecast experiments of thirteen rainfall cases during the period from June to August 2016. A comparative study is then conducted based on assimilation and forecast results with and without ZTD data to evaluate the effect of assimilating ZTD data into the RMNPSNC. The results are summarized as follows. (1) The errors of ZTD observations diagnosed by the Des method are relatively reasonable, and the differences of diagnostic values between stations are large, indicating the necessity of diagnosing observation errors station by station. (2) Assimilation of the ZTD data improves the forecast of the intensity and distribution of heavy rain and the forecasts of temperature, humidity and wind are closer to observations. The analysis and forecast effects of the Des scheme are better than that of the traditional scheme. (3) For the heavy rain process in Northeast China on 25 July 2016, the assimilation of ZTD effectively increases the initial humidity field, improves the content and spatial distribution of hydrometeors at the initial hours of integration, corrects the failed precipitation forecast in the east of Liaoning province, and improves temperature and wind forecasts due to realistic precipitation feedback. The diagnostic ZTD observation errors obtained by the Des method are more reasonable than that by traditional method. Therefore, the assimilation and forecast can be improved by using the Des method. Progress in the forecast of rain, temperature, humidity and wind can be made through assimilating the ZTD.