To further improve the precision of the Continuously Operating Reference Stations (CORS) coordinate solution, high-precision data processing software was used to process the multi-year observation data of 13 CORS in Beijing, and the original coordinate time series and coordinate residual time series were obtained. Maximum likelihood estimation combined with power spectrum estimation was used to analyze the noise category and the best model of the CORS. The results show that the noise components in the north, east, and up (NEU) directions of the time series include white noise and flicker noise, among which the noise components in the N direction of seven CORS include random walk noise, the noise components in the E direction of one station include random walk noise, and the noise components in the U direction of three CORS include random walk noise. The best noise model of the Beijing CORS base station coordinate time series is the white noise (WN)+flicker noise (FN)+random walk noise (RWN) combined model.