Abstract: The un-differenced GPS observations are usually adopted by precise point positioning (PPP), of which the initial phase biases (IPBs) underlying the phase observables are inseparable with the integer ambiguities, leading to ambiguity-float PPP-based solutions. Currently, the satellite IPBs as retrieved from regional or global GPS networks can be used for correcting PPP user’s phase observable and restoring the integer nature of the ambiguities. After successful ambiguity resolution, improving accuracy and convergence behavior of PPP-based solutions can be expected. The commonly used methods for estimating satellite IPBs include estimation of decoupled satellite clock or fractional phase biases. Starting from the original GPS observations, the full-rank mathematical model for satellite IPBs estimation is derived and used for comparatively analysis of the characteristics and implementation of both methods. It is concluded from the analysis that full utilization of information from GPS observations can be expected for the former method, and the process of generating decoupled satellite clocks is consistent with the standard satellite clocks estimation, but doesn’t make good use of the stable temporal behaviors of satellite IPBs. In contrast, the latter method doesn’t fully account for the stochastic correlations between the linear combinations of GPS observations, which would lead to a sub-optimal satellite IPBs, besides, the real-time implementation of this method is more troublesome than the former, and the GPS code observations of high-accuracy are always required. The strategy presented in this paper can avoid the shortcomings in both methods, and reasonable constraints due to the stability of several unknowns can be easily imposed upon during parameter estimation to generate more reliable satellite IPBs.