Abstract: The CGCS2000 (China Geodetic Coordinate System 2000) was first released on July 1 in 2008. After that, all location-based GNSS (global navigation satellite system) applications are required to be processed and corrected to CGCS2000. As a matter of fact, big differences exist when different implementation strategies or schemes are adopted, such as reference station selection rules, area partition scheme, and different epoch correction methods et al. All of the above-mentioned factors lead to different results. Some of the differences even reach decimeter level. For instance, the coordinate differences based on plate movement correction with velocity field and the quasi-stable adjustment under CGCS2000 may reach decimeter level. The reason why there exists so big differences in coordinate solutions is the fact that the solution depends on a person's different understanding in various choices by the software. In another word, GNSS data processing is lack of detailed guidance from technical standard. In this paper, a statistic method is proposed. Firstly, supervised clustering is proposed as the rule for reference station selection, Secondly a so-called Spacing Zoning method is used for area partition which is a necessary step facing at a large network. And thirdly a plate movement correction method is applied for correcting position from current epoch to epoch 2000. The station coordinate based on the spacing partition scheme is better than that of the regional dividing scheme, and the coordinate precision of the three-dimensional direction is better than 2 mm. Through the reasonable optimization strategy, the root mean squares of the velocity derived in X, Y, Z directions are reduced from 0.92, 0.72, 0.97 mm/a to 0.19, 0.45, 0.32 mm/a respectively. All of these methods used here obviously have improved the CGCS2000 frame maintenance accuracy.