In this study, we evaluated the travel time of S-wave between the vertical array stations based on seismic interferometry, focusing on the difference in transfer function due to two horizontal components at the KiK-net Mashiki station (KMMH16). At that time, we surveyed the differences by back azimuth (BAZ) and the polarization direction of seismic waves. Furthermore, we expanded the survey to all KiK-net stations in the Kyushu district, to confirm whether the phenomena seen at KMMH16 is specific to this location. The result shows that the difference by the polarization direction in the travel time was larger than the difference by the BAZ. This result suggests that the difference in transfer function at KMMH16 were affected by the anisotropy of the S-wave velocity. We evaluated the leading S-wave polarization directions (LSPDs) and the strength of anisotropy (ΔV) for all KiK-net stations in the Kyushu district. The LSPDs roughly correspond to the results of previous studies. The LSPDs in the forearc area are nearly perpendicular to the crustal deformation whereas those in the back-arc area are nearly parallel to it. This characteristic is similar to one found by Nakajima and Hasegawa (2008) in the Tohoku district. We examined the change in anisotropy before and after the Kumamoto earthquake at two stations, KMMH16 and KMMH14 that are located near the source region. The changes in the LSPD and the ΔV before and after the earthquake were not notable. At stations that observed weak anisotropy, transfer functions of two horizontal components show similar shape. At stations that observed strong anisotropy, however, the shape of the transfer function differs greatly, depending on the horizontal direction. This suggests that an evaluation of site amplification using a single velocity model may reduce the reproducibility of ground motions.