The accurate estimation of seismometer orientation is critical for seismological studies relying on three-component seismic records. In this study, we estimated the seismometer orientations of 352 broadband seismic stations from ChinArray phase I, which located in the southern section of North-South Seismic Zone during 2011-2013. Two independent polarization analysis approaches are adopted using both the teleseismic P-wave and Rayleigh surface wave. In total, the P waves from 696 teleseismic events with epicentral distance 30-90 degrees and Rayleigh waves from 1928 events and events from 10 to 170 degrees were used respectively for orientation estimation. The results from the two methods show a comparatively good consistency. Among 352 stations, there are 4 stations having records of less than 10 events , which are not evaluated. For the others, 281 stations have orientation deviations of less than 10 degrees from the north, 21 stations are between 10 and 20 degrees, and the remaining stations have much larger deviations or instrument failure. The results are similar to those of previous researchers from teleseismic P-waves analysis, but there are 7 stations inconsistent with their results showing deviations larger than 5 degrees. Further analysis shows the orientations of these stations are time dependent and change during the entire operation period. Furthermore, anormal gains between 3-component channels can produce bias on the orientation estimation, which can be distinguished by jointly inspecting the noise power spectrum. The approaches and results of current study provide orientation correction information for ChinArray phase I and can also be applied for other mobile stations for better data quality evaluations.