【Objective】 The study is aimed to explore the law and influencing factors of soil CO2 emission under three typical vegetation (Xanthocarp woodland, barren grass land and seabuckthorn shrub) cover in Loess hilly re‐ gion, to compare differences of CO2 emission in soil covered by different vegetation types. It provides the basis for se‐ lecting suitable vegetation for returning farmland to forest and grassland to reduce soil CO2 emission in this area. 【Method】 Based on the DNDC model, combined with the actual soil CO2 emission flux measured, the suitability of the model to simulate soil CO2 emission under different vegetation cover was tested, and its sensitivity was analyzed. 【Result】 The results showed that the seasonal variation trend of forest soil CO2 emission in Loess hilly region had a single‐peak curve. The peak value appeared between early July and late August, and the minimum value appeared be‐ tween late December and early January of the next year. There was a significant positive correlation between soil CO2 flux and soil surface temperature (5 cm) and soil water content (P<0. 01). Soil temperature, moisture content were the key factors affecting CO2 emissions. The average annual CO2 emission flux of the soil in the xanthocarp forest was the lowest, which was 40. 785% and 40. 835% lower than those in the grassy land and seabuckthorn shrub, respec‐ tively. The simulation results of soil CO2 emission among different vegetation types by DNDC model were consistent with the measured results. The R=0. 928(P<0. 01) in the Xanthocarp fruit forest land, the R=0. 932(P<0. 01) in the Barren grassland, and R=0. 983(P<0. 01)in the Seabuckthorn shrub land. The results showed that the model could be used to simulate the CO2 emission of forest land in this area.【Conclusion】 The model sensitivity analysis showed that soil surface SOC content was the sensitive factor to simulate the CO2 emission of three typical vegetation-covered soils in this region,and the change degree of SOC content had the greatest influence on the model simulation results.