Abstract The misorientation of three-component seismometers restricts the application of relevant seismic experiments such as ocean-bottom seismometer (OBS) arrays. Previous orientation determination relied on an assumption that the propagation azimuth of seismic waves follows the great-circle path (GCP) azimuth. This assumption may yield systematic errors in the estimated orientation, particularly when the ray paths are bent laterally due to velocity heterogeneity in the Earth. Here, we develop a new method for unbiasedly estimating the horizontal orientations of seismic sensors and apply this method to the Blanco transform fault OBS experiment. We first retrieve the orientations relative to the propagation azimuths from the recorded Rayleigh and P waveforms, and then determine the geographic north orientations by calculating the propagation azimuths via an Eikonal-equation-based phase-tracking method that theoretically accounts for the effect of ray bending. Synthetics test validates that the phase-tracking method can retrieve unbiased propagation azimuths of seismic waves. The final results derived from Rayleigh- and P-wave polarization analyses with the respective phase-tracked propagation azimuths are more consistent and the orientation errors are smaller, indicating the robustness and accuracy of this method. Comparing the orientations from our phase-tracking method to those from the GCP assumption, the deviation can reach up to 8° between these two techniques in the study region. Subsequently, when orientations of the synthetics modeled from three-dimensional elastic waveform simulation are deviated according to the GCP-predicted orientations, we find nonnegligible bias in the phase and amplitude measurements that could reduce the accuracy and resolution of following inversion, which indicates the significance of our phase-tracking method in accurate orientation of OBS arrays as well as inland seismic experiments.