Due to the existence of the aerosol, the traditional method of measuring atmospheric temperature by using Rayleigh scattering technique has limitations in the low altitude. A pure rotational Raman lidar to get tropospheric temperature profiles is built. We carried out the atmospheric temperature observation in Beijing for two months. The atmospheric temperature profile was retrieved using the observed rotational Raman scattering signals. The effect of smooth window, calibration range and calibration constant on the retrieval precision of the atmospheric temperature was evaluated and analyzed. The results show that with the increase of smooth window, the mean absolute deviation between the lidar and radiosonde firstly decreases and then increases; in order to remove effectively the effect of random error in the return signals, while maintaining the fine vertical structure of temperature profile, it is better to choose the range between 600 and 1 200 m for smooth window. When calibration range is different, the mean absolute deviation between the lidar and radiosonde is varied, the relative variation of the deviation is about 0.07 K. When both calibration constant a and b increase or decrease, the mean deviation between the lidar and radiosonde increases; when one increases and another decreases, the mean deviation has a tendency to cancel each other out. The variance probability of a or b is not equal, and the variance of a and b is always contrary in the sign; the mean deviation is not sensitive to variance of a or b, and it is sensitive to the whole variance of a and b, about 91.7% of the mean deviation is in the range between -3 and 3 K. These results provide the theoretical basis for the selection of smooth window and calibration range in pure rotational Raman lidar data retrieval, and the reference for the error of actual temperature inversion result caused by lidar calibration constant.