Abstract The time-varying gravity field obtained from the GRACE gravity satellite has significant north-south stripe errors, which greatly mask the true gravity field signal, necessitating the need for filtering processing. Based on the GRACE correlation filtering theory and the evaluation basis of the optimal signal-to-noise ratio, we analyze in detail the changes of the signal-to-noise ratio and the Gaussian weight coefficient of different Gaussian filtering half-diameter under each order. Based on this, we propose a mixed radius Gaussian filtering method, which aims to optimize the Gaussian filter weight coefficients by adjusting the filter radii at each order. The results indicate that when only the combination of 400 km classical Gaussian filtering, de-correlation filtering and 300 km classical Gaussian filtering has the best signal-to-noise ratio, two-step 400 km mixed radius Gaussian filtering, de-correlation filtering and 300 km mixed radius Gaussian filtering can further improve the signal-to-noise ratio. Additionally, these methods also effectively address the signal leakage caused by Gaussian filtering.