Utilizing soil temperature data collected at a bare ground soil over the Loess Plateau during the period from 22 to 26 July 2005，the soil thermal diffusivity is calculated by using two algorithms (i.e., thermal conduction algorithm combined with mathematical fitting method and thermal conduction-convection algorithm) in order to better understand the physical property and procedures of the soil over the Loess Plateau quantitatively and to form a basis of improvements of soil temperature diagnosis in land surface models. Soil thermal diffusivity is calculated by the two methods mentioned above, and the sum of vertical gradient of soil thermal diffusivity and water flux density is calculated by the thermal conduction-convection algorithm to range from 0.80×10-6 to 2.43×10-6 m/s for the soil layer of 0.05－0.10 m. Taking the soil layer at the depth of 0.05 m as the upper boundary, the temperature for the soil layer at the depth of 0.10 m is modeled by means of the two methods mentioned above. It is found that for the thermal conduction algorithm, owing to the fact that it takes into account thermal conduction alone and considers neither the soil water movement nor the soil vertical heterogeneity, both the diurnal amplitude and the delay of phase are overestimated. For the thermal conduction-convection algorithm, the diurnal amplitude and phase can be well estimated. The bias, standard error and normalized standard error of the modeled value against the observation are respectively 0.19 K, 0.18 K and 0.08% during the daytime (0800－2000 Beijing Time).