To address the issue of attitude determination for the autonomous navigation of the unmanned vehicles, an approach to define the pitch and roll angle of the vehicle by means of the stable distribution of skylight polarization pattern was proposed. Firstly, based on the Rayleigh scattering theory, we establish the theoretical and practical detected patterns of the polarized skylight. Secondly, through optimization, we obtain the data of the astronomical marker—the location of the sun based on the polarization data detected. Then, with respect to the relatively consistent space-time relation between the sun and the zenith, the location of the zenith is thus figured out. Finally, the coordinate of the zenith is adopted to establish the attitude rotation matrix and by which to work out the pitch and roll angles. In addition, by setting up a simulation vehicle platform to acquire the theoretical and practical polarization data, the effectiveness of the approach is tested. The calculated root-mean-square errors of the pitch angle are 0.002° and 0.0031° respectively for the theoretical and practical simulation test, and errors of the roll angle are 0.0236° and 0.0227° respectively. The results show that the approach to attitude determination through polarization detecting has advantages in easy handling and high accuracy.