The horizontal orientation estimation of an ocean‐bottom seismometer (OBS) is critical for providing reliable data for seismological studies. In this article, we applied three independent polarization analysis methods to estimate the absolute horizontal orientation of a pilot shallow‐water OBS array in Bohai Sea, China. The OBS array was deployed in the autumn of 2018 and comprised 32 broadband seismometers lasting around 1 yr, aimed at investigating the geodynamics of North China craton and adjacent areas. The timings of the data were corrected through the correlation analysis of regional and teleseismic earthquake events, before the polarization analysis. Polarization analysis using P wave and Rayleigh wave, from both natural earthquakes and noise cross correlations, was performed. In total, 28 out of 29 stations obtained the effective estimates combining the three methods, of which 11 stations showed relatively robust results with variations well below 10° among the three methods. However, the superiority of one method over the others is not obvious. As there is a trade‐off between the accepted final estimates and the number of qualified measurements defined by parameters such as the correlation coefficient threshold, window length, and filtering options, we investigated these effects using 15 different groups of parameters for the three methods, and the deviation statistics showed a distribution with the root mean square deviation of 9.2° for the whole array.