The vertical distribution of the atmospheric temperature in the troposphere is directly related to the meteorological phenomena and the diffusion of atmospheric pollutants. It has been the major parameters observed by the meteorological department and the environment sector. The lidar technology has become an effective method to detect the vertical distribution and time variation of the atmospheric temperature in the troposphere. Lower tropospheric temperature profile measured by lidar using the Rayleigh and vibrational Raman scattering can′t be obtained accurately due to the abundant aerosols. Using N2 and O2 molecular pure rotational Raman scattering signal, the lower tropospheric temperature profile can be obtained without the influence of lower tropospheric aerosol theoretically. The main difficulty of the rotational Raman lidar is design and mechanics of the receiving spectroscopic system. In domestic research, most of the lidar systems use the spectroscopic technique based on double grating spectrometer. The technique based on the interference filters was introduced in this paper and used in our lidar system for observation of the tropospheric temperature. This technique had more efficiency and suppression in separation of the pure Raman signals from Mie signals. Furthermore, the system′s sensitivity can be optimized by selecting the tilting angle of the filters. The experiment was based on the project of "Formation Mechanism and Control Strategies of Haze in China" carried out by the research groups of the Chinese Academy of Sciences. Our lidar system was moved to the super atmospheric observatory in University of Chinese Academy of Sciences on November 2014 and the experiment were taken during the APEC conference. The energy of laser in ultraviolet was about 200 mJ, the frequency was 20 Hz, the laser pulse number was 5 000, the spatial resolution was 7.5 m. The experimental result shows that the statistical error between lidar and radiosonde are less than 1.5 K with the range up to 10 km and the statistical error of less than 1 K is obtained up to 7.5 km height in the clean night. Larger differences of about 3 K at the region above the thin cloud may be caused by the strong elastic backscatter signal and the statistical error of less than 1 K is obtained up to 4.8 km height.