It is important to diagnose and analyze 2 m temperature prediction by CMA-MESO system with 3 km resolution in winter for the Winter Olympics meteorological service and CMA-MESO model system development. The 2 m temperature prediction data from December 2020 to February 2021 are selected and diagnosed. The results show that the daily low temperature prediction is better, but the high temperature prediction is poor. The prediction effect of the model on the heating process in 00-06 UTC is better than that of the cooling process in 06-21 UTC. When the 2 m temperature prediction is corrected by the moving-biweight average method, the correction results show that the 2 m temperature prediction deviation is mainly systematic deviation. The RMSE and standard deviation are reduced, especially in areas with large deviations before correction. According to the large difference of standard deviation before and after temperature correction, individual cases in North China, East China and Southwest China and the continuous test in North China are selected for wave spectrum analysis. It is found that the power spectrum energy gradually increases with the scale. There is a certain correspondence between the 2 m temperature prediction deviation and the power spectrum information of different scales. When value of power spectrum energy is very small or abnormally large, the difference between 2 m temperature prediction and observation is significant.