Abstract: Based on the correlations between the noise data of the horizontal and the vertical components, we can remove the tilt noise from the vertical component by using the transfer function of the horizontal-to-vertical component. Using the data from the 2019−2020 Pankun OBS South China Sea test, this paper describes the theory and process of this method. It compares the seismograms before and after removing the tilt noise and highlights the improvement of the dispersions of Rayleigh surface waves. The results show that removing tilt noise can improve the signal-to-noise ratio of seismograms for the OBS data in the low-frequency range, resulting in surface waves more conducive to imaging the deep structure of the oceanic lithosphere. This study also shows that although the tilt angle （1.0°） of the seismometer of Pankun OBS is much smaller than the tolerance of the tilt angle of the instrument （2.5°）, the bottom currents still generate a significant level of tilt noise on OBS data. Therefore, the leveling system of the instrument is a crucial component affecting the OBS data quality.