【Objective】Mean annual soil temperature is an important physical property of soil and of great significance to regional agricultural production. The purpose of this study is to explore regional characteristics of mean annual soil temperature relative to soil depth in China, and define an applicable method for accurate estimation of mean annual soil temperature. 【Method】Ground meteorological data accumulated in the 895 meteorological stations in China during the period from 1980 to 2010 were collated and mean annual soil temperature analyzed based on zoning by Grade I standard farming system in China. Two methods were tested for estimation of mean annual soil temperature in China. The first one is a multiple regression equation established on the basis of SPSS Statistics 17.0 for regression of mean annual soil temperature with mean annual air temperature, latitude, longitude and altitude, and the second one is to estimate mean annual soil temperature based on the mean annual air temperatures and regional mean air temperature already available.【Result】 Results show that the mean annual soil temperature in various soil layers 5 to 40cm in depth in all the regions was higher than the mean annual air temperature monitored at the meteorological stations of corresponding regions. In the same meteorological zone, regardless of meteorological stations, mean annual soil temperature varied with soil depth within 5 ~ 40cm, but the variation did follow the same trend. Mean annual soil and air temperatures differed in the range of ≤0.1℃in all the soil layers from 5 to 20cm in depth, but varied more,significantly in the soil layers from 20 to 40cm in depth, and the difference in some individuals zones reached 0.4℃. Comparison between meteorological zones shows that the difference between mean annual soil and air temperatures varied regionally, that is, declining first and then rising from north to south, and gradually increasing from east to west; in some zones the difference varied in the range of 1.4～3.9℃at 20 cm in soil depth and in the range of1.1～4.3℃ at 40cm in soil depth. Compared with the estimation model based on zoning by Grade I standard farming system, the regression equation was higher in accuracy, but it was not so significant in the Qinghai-Tibet Plateau, Inner Mongolia and zones alongside the Great Wall. Analysis of the data of mean annual soil temperatures in soil layers at 5, 10, 15, 20 and 40cm in depth at all the 262 meteorological stations that had kept complete data shows that in the soil layers 5～40cm in depth, mean annual soil temperature varied in the range of ≤0.1℃ at 5cm and much less in the soil layers 20～40cm in depth. Based on a single estimation site, the estimated mean soil temperature in the soil layer at 50cm differed from the value at 40cm to the extreme by ≤0.4℃.【Conclusion】The estimation of mean annual soil temperature using the mean annual soil and air temperature difference method is quite low in accuracy and the method is recommended for use only in the Middle and Lower Reaches of the Yangtze River, Jiangnan region, Sichuan Basin and Yunnan-Guizhou Plateau, and the regression equation method is quite high in accuracy in all the zones of Grade I standard farming system, except for Qinghai-Tibet Plateau, Inner Mongolia and zones alongside the Great Wall. In farming zones where the distribution of meteorological stations is dense and little in elevation variation, the regression equation based model is the more higher in accuracy with little variation in mean annual soil temperature in the soil layer 40~50cm in depth. Therefore, it is advisable to use the soil temperature at 40cm to replace that at 50cm in the prevailing “Chinese Soil Taxonomy”.