Abstract: Besides GPS and GLONASS, Galileo and BDS are developing fast, a multi-Global Navigation Satellite System(multi-GNSS) age is emerging, which brings a great opportunity to improve the quality of Positioning, Navigation and timing(PNT) by combining multi-GNSS and multi-frequency data. The data quality of GNSS itself is one of the preconditions for achieving high-precision results, and it is also the key to construction of stochastic model in multi-GNSS precise positioning. In order to avoid the negative effects of Code Division Multiple Access(CDMA) and Frequency Division Multiple Access(FDMA) mechanism for different GNSS systems, this paper used Geometry-free and M-W combination method to compare and analyze carrier phase single-difference(SD) residual time series for four systems where data of a zero baseline measured in Curtin University are adopted. Meanwhile, elevation-dependent stochastic models are obtained for each GNSS system based on the sinusoidal model and the exponential model. The experimental results show that the SD residual series perform different characteristics in different frequencies and systems. The precision of L1 and L2 observations in Galileo system is quite similar and about 0.9mm; however, precision of L2 observations in other systems is even worse than that of L1. The fitting results of elevation-dependent model show that the sinusoidal model and exponential model have a good consistency with precision of L1 and L2 observations of GPS and Galileo, while the sinusoidal model has a slightly worse fitness in BDS system, and using sinusoidal model to evaluate the accuracy of L2 observations in GLONASS system is inappropriate.