Abstract: Principal component analysis method is used to determine the contribution of elevation angle and signal to noise ratio (SNR) in observation noise, and a refined Global Navigation Satellite System (GNSS) stochastic model is established based on the analysis results. The performance of the refined stochastic model is verified by using precision point positioning (PPP). It shows that the refined stochastic model leads to better positioning results in high latitude areas than traditional model that only takes into account elevation angle or SNR. The refined stochastic model is about 30% more accurate than elevation angle model, and about 20% better than SNR model. The accuracy of refined stochastic model improves most obvious in the zenith direction, and the improvements are about 38% and 24% with respect to the results of elevation angle model and SNR model, respectively. This study indicates that our new refined stochastic model is advantage to high-precision positioning accuracy in high latitude areas.