Non-photosynthetic vegetation (NPV) is a vital component of terrestrial ecosystem, as well as a critical hotspot in global carbon and nitrogen studies. In addition, its coverage in the source area of wind-blown sand in northern China indicates the intensity of wind-blown sand activities. As revealed from recently conducted studies, cellulose absorption index (CAI) was highlighted to be capable of more effectively reflecting the fractional cover of non-photosynthetic vegetation (f_NPV). However, there was no corresponding band (2000-2050 nm, 2080-2130 nm, and 2100-2240 nm) for CAI calculation in the 36 bands of MODIS data. Subsequently, under the introduction of dead fuel index (DFI), it can be employed directly to estimate f_NPV, whereas a question is whether DFI accurately reflect the content of NPV cellulose? Is it necessary to establish the relationship between CAI and DFI to achieve more accurate estimation of f_NPV? It has always been neglected by researchers. Therefore, this paper intended to study the distribution of f_NPV in various grassland, and compared the difference between direct estimation of f_NPV by DFI and indirect estimation of f_NPV by CAI, the influence of bare land on the estimation results were analyzed in order to explain the spatial heterogeneity characteristics of f_NPV in different grasslands. We analyzed the NPV spectral characteristics of different habitats NPV (herb, semi-shrub and shrub) in sandy grassland of Ordos and constructed the application model of DFI, CAI and f_NPV based on the ground hyper-spectral data measured by analytical spectral devices, and then validated in different grassland areas. The results showed that the NPV reflectance had similar trends of herb, semi-shrub, shrub and mixed NPV. There was a significantly positive correlation between DFI and CAI, CAI and f_NPV, DFI and f_NPV at the level of P<0.001, and DFI could better reflect the change of f_NPV. But the f_NPV value directly calculated by DFI was larger than the f_NPV that calculated indirectly by CAI, the average overestimation was 3.11% when eliminating bare ground. The difference between two values in typical grassland areas was 0.03%. In addition, the NPV should not be treated as a homogeneous material, and bare ground interference should be eliminated when performing f_NPV estimation.