In order to find an effective and precise method for estimating aboveground biomass and grassland biomass of the Loess Plateau, field surveys at the individual and population levels were conducted, and an estimation model was established. The model included the composite factor CH (product of vegetation coverage and height) as the independent variables, and regression analysis was used to estimate the grassland biomass of the Loess Plateau in mid-August of 2014. Simultaneously, the accuracy of the estimation models of aboveground biomass were assessed using the leave-one-out analysis method combined with the correlation coefficient (r), average absolute value of relative error (RMA), total relative error (RS), and correction factor. The results showed that all estimation models of aboveground biomass were linear and exponential functions at both the individual and population levels in the grassland of the Loess Plateau. The results of accuracy testing of the estimation model showed that the relationship between the estimated and measured biomass values were well correlated at the individual level, and reached a significant level (P < 0.05) with almost all r values greater than 0.6, RS values less than 10%, RMA less than 30%, and the correction coefficient close to 1. However, at the population level, although the correlation between the estimated and measured biomass values was significant (P < 0.05), the RMA of most species was greater than 30%, the RS values were greater than 10%, and the measured values were less than the estimated values with correction coefficients deviating from 1. Together, these results showed that the accuracy of estimation models of aboveground biomass established at the individual level is better than for those established at the population level in the Loess Plateau grassland.