This study modeled the spatial and temporal distribution of soil temperatures in the Xinjiang region of China, using atmospheric surface forcing data in the China Meteorological Administration Land Data Assimilation System (CLDAS, NMIC of China Meteorological Administration) to drive the Community Land Model (CLM3.5, National Center of Atmospheric Research USA) for hourly off-line simulations (from 2009 to 2012). To verify the CLM3.5 simulated soil temperatures, data from national automatic soil-temperature stations (105 in the Xinjiang region) were used at three soil layers (5 cm, 20 cm, and 80 cm). For monthly variation, simulated top layer (5cm) soil temperatures differed substantially from measured values, with the largest difference (±5℃) reaching the maximum in July each year. The difference (±3℃) between modeled and observed soil temperatures at the second layer (20 cm) reached the maximum in July for all years, whereas for the third layer (80 cm), simulated annual July soil temperatures were in accordance with the observed values. The large discrepancies in July soil temperatures in the top surface layers can be explained by the drastic surface temperature changes in the Xinjiang region during that month. With day-time temperatures that can reach above 30℃, combined with large diurnal temperature differences, it becomes very difficult to accurately capture surface temperature variation by using the model. In contrast, in January and December, the 80 cm soil depth simulations were less accurate than the results of simulations at the first two soil layers. Furthermore, simulated values of soil temperature at the top two layers (5 cm and 20 cm) did not fit well with observed values for the summer and autumn. However, similar to monthly variation, the daily variation in modeled soil temperature at 80 cm showed a bad fit with observed data in January and December, whereas the fit was good in other periods. For hourly variation at 5 cm soil depth, the simulated soil temperature values were higher than the observed ones from January to April and September to November between 03 UTC and 21 UTC the next day. In contrast, simulated results were slightly lower than the observed values between 21 UTC and 00 UTC the next day for this same layer and period. From May to August of every year, day-time simulated values are slightly lower, reaching a maximum at 09UTC. At 20 cm depth, simulated soil temperature had smaller deviations (between -1℃ and 1℃) for most months, and the daily maximum occurred at 12UTC, which is earlier than the observed values. At the 80 cm soil layer, little daily variation was simulated or observed in soil temperatures, giving this soil layer hardly any influence on the overall daily variation. In the Xinjiang region, the diurnal temperature difference is large from May to August and September to November, which can explain why the upper two soil layers show differences between modeled and observed hourly soil temperatures. Soil temperature at deep soil layers, however, will vary less with temperature differences, giving this layer a better fit than the other two. Overall, this study shows that the CLM3.5 model forced by a CLDAS driving field can simulate the multi-year spatial and temporal distribution of average soil temperatures in Xinjiang region precisely. It furthermore showed that this method could simulate and reflect the hourly, daily, monthly, and yearly patterns of soil temperature in the Xinjiang region. Finally, the poor simulation of temperatures at the surface layer may be caused by the parameterization scheme of the surface parameters in this model, which will be addressed and corrected in a later phase.