Abstract: Through the comparative analysis of the forecast results of the regional numerical models (CMA-MESO-3KM, SW3KM, SW9KM）and the actual data of meteorological observation stations of daily 2m temperature ( including daily maximum temperature and minimum temperature ) from March to December 2021, the forecast performance of these three numerical models for 2 m temperature in Sichuan Province was tested and evaluated. The results showed that: (1) The model forecasts for the minimum temperature were better than those of the maximum temperature. SW3KM model was slightly better than the SW9KM and CMA-MESO-3KM models in predicting 2 m temperature from March to October, the highest and lowest temperature accuracy rates of SW3KM model were higher than those of SW9KM model, up to 13.4% and 31.9% respectively, and up to 18.8% higher than that of CMA-MESO-3KM model. (2) The errors of temperature forecast showed daily variation. The error of the three models gradually decreased from early morning to morning and gradually increased from afternoon to evening, the morning temperatures of the SW3KM and SW9KM models were higher than the observation, the temperature in other forecast periods are lower than the observation, and the forecasting temperatures of CMA-MESO-3KM model were lower than the real situation. (3) The temperature accuracy was closely related to the altitude, the forecast accuracy decreased with the the increase of altitude, and the systematic deviation increased. (4) The CMA-MESO-3KM model had a higher temperature forecast in the cold season and a lower forecast in the warm season, and the forecast from 12:00 p.m. was slightly better than that from 00:00 a.m.. The temperature forecast of SW3KM and SW9KM models above 20°C startiing at 00:00 a.m. was lower than the real situation, and the temperatures forecast below 20°C was higher than the real situation, while the temperature forecast below 30°C starting at 12:00 p.m. was lower than the real situation , but the temperature error starting at 00:00 a.m. was less than 12:00 p.m.