Abstract: In this study, based on the Rayleigh lidar echo photon signal of the middle atmosphere, the optimal estimation method is used to retrieve the atmospheric temperature profile. The forward model is constructed by using the Rayleigh lidar equation, and the covariance matrix of the measured signal is determined according to the Poisson counting characteristics of the photodetector. The temperature profile of the atmospheric model is selected as the prior state information, on the basis of which the cost function is determined. Finally, the Levenberg-Marquardt optimization algorithm is used to optimize the cost function. The average kernel matrix is used to evaluate the contribution of real information to the inversion results, and the uncertainty of the inversion results is calculated. The results show that the forward model can correctly describe the real physical process of Rayleigh lidar detecting atmosphere. In the area where the signal-to-noise ratio of the echo photon signal is high, the real information accounts for the main contribution to the inversion results, and the vertical resolution is small. The uncertainty of temperature inversion in the range of height below 90km is between 0K and 10K.