The paper took a repeated two-factor experimental design to explore the response characteristics and its variation of different grazing system and soil depth on underground biomass. Combined with regression model, it established fitting and forecasting functions of community underground biomass variation so as to pro- vide a theoretical guidance and basis for the management of degraded desert grassland and restoration of grazing e- cosystem. The results showed that underground biomass mainly concentrated in the surface layer （0 - 40 cm）, and sharply decreased with increased soil depth, the effects of different grazing systems on underground biomass limited within 0 - 40 cm soil layers （P〈0.05）. The total underground biomass in rotational grazing area （824 g/m^3） was significantly higher than that in continuous grazing （551 g/m^3） and enclosed grazing （520 g/m^3） （P 0.05）. According to the prediction model design, the total biomass can get more than 90% of the information for actual measured underground biomass, which provides a theoretical basis to impact of differences among different grazing systems on underground biomass and its variation with soil depth, and also provide theoretical guidance for grassland management and sustainable grassland utilization.