The accuracy of stochastic model directly affects the precision and reliability of GNSS precision positioning.With 10-day observation data of CUT0,CUT2 and CUTA stations in MEGX observation network,the residuals time series of GNSS were solved by single-difference model using short baseline method.According to the statistically observed noises changing with the elevation angle,a refined stochastic model is established via the least-square method.The results show that the range of pseudo-range observation precision lies in 20-50 cm,30-50 cm and 42-63 cm for Galileo,BDS and GPS,respectively;and the carrier phase observation precision falls within the scope of 1-3 mm for Galileo and BDS,and on 3 mm for GPS.For the four frequencies of Galileo,E5 is most accurate in both pseudo-range observation and carrier phase;the accuracies of B1,B2,B3 of BDS are roughly the same;the L2 of GPS is more accurate in pseudo-range observation with precision of 42 cm than L1 of 63 cm,and both L1 and L2 have carrier phase precision of 3 mm.From the above precision estimations based on single-difference model,it can be concluded that in case of short baseline,BDS/GPS/Galileo have roughly same precision in carrier phase observation values,while Galileo E5 outperforms the other satellite systems in pseudo-range observations.Residual analysis of satellite observations provides a basis for the establishment of stochastic models in positioning.