Based on China Meteorological Administration (CMA) ensemble numerical forecast products from the TIGGE data, the probability forecast experiments of the surface temperature are carried out. The results show that by using the methods of Downscaling (D), Systemic deviation (S), and combination of Downscaling and Systemic deviation (D S), through the Brier Score (BS) and Relative Operating Characteristic (ROC) tests, forecast skills have great improvement in 24〖CD*2〗 to 240〖CD*2〗hour forecasts. Particularly, the method D S is much better than methods D and S. Calculations of ranked probability score (RPS) indicate that the result of the method D S is greater than those of methods D and S before 168 hour forecasts； after 168 hours, the method D S is much less than that of the method S but better than the of the method D. Averagely, the probability forecasts of three methods are all skillful for surface temperature. The skill of probability forecast for surface temperature before 168 hours is greater than the skill for the later 168 hours.