In order to achieve high signal-to-noise ratio by using small laser energy and telescope aperture, we present a polarization filter in high-spectral-resolution lidar (HSRL) for the measurement of atmospheric temperature. Compared with the filter method in a traditional HSRL in which the intensity of the return signal is split into the different transmission channel of a discriminator, the advantage of this filter system is that the intensity of the return signal is fully utilized for each discriminator channel, and the return signal changes the polarization state of the light without loss of intensity when it is incident on the two Rayleigh channels. In addition, the daytime detection capability of HSRL is improved by using a polarization optical scheme to suppress the solar background light. The advantages of the polarization filter are proven by the theoretical calculations using the Stokes vector and a Mueller matrix. In detection experiments of atmospheric temperature, the detection height is 4 km at night and 2.5 km during the day by using the pulsed energy of 50 mJ and telescope diameter of 250 mm. The results are in good agreement with the data detected by radiosonde.