This study quantitatively evaluates the performance of 2 m air temperature forecasts from the ECMWF model over Guizhou Province to enhance its operational application. Based on 24 h maximum and minimum temperature forecasts within 168 h lead time from 2020 to 2023, comprehensive assessment is conducted on the temperature forecast performance across the entire province, considering typical weather events and varying altitudes. The results indicate that: (1) The mean forecast accuracy and root mean square error (RMSE) for both 24 h maximum and minimum temperature forecasts within 168 h lead time exhibit significant regional, inter_annual and monthly variations. The accuracy of daily minimum temperature forecasts is consistently 30% - 40% higher than that of maximum temperature forecasts, and has greater stability. (2) On a monthly scale, the accuracy of daily maximum temperature forecasts shows a multi_peak distribution, with peaks appearing in June and October. However, the accuracy for daily minimum temperatures follows a single_peak pattern, concentrated in July. (3) During cold wave and sharp cooling events, the forecast accuracy of daily maximum (minimum) temperature remains relatively stable from 168 h to 72 h ahead. However, the accuracy markedly decreases within the 48 - 24 h lead time, accompanied by a sharp increase in the average RMSE to 6. 0 - 10. 6 ℃. (4) The forecast quality varies with altitude in a single_peak distribution, achieving optimum performance at elevations between 1000 m and 1500 m, but degrading towards both higher and lower altitudes. This study elucidates the quantitative characteristics of ECMWF temperature forecast errors in the complex terrain of Guizhou, and could be a scientific basis for optimizing its operational application.