129 loess soil samples taken from the field in Qian County of Shaanxi Province in 2014 were chosen as objects to build the inversion model between soil moisture content and spectra. The spectra and gravimetric moisture content of soil samples were measured during the process of soil air drying, and the relationship between spectra and soil moisture content was analyzed. The spectral predictive models of soil moisture content were established by using the linear regression and exponential analysis. Results showed that the biggest correlation coefficients and absorption depth bands located in 570, 1460, 1960nm and 490, 1460, 1960nm in the region of 400~1340, 1460~1790, 1960~2390nm, respectively. The linear relationship between spectral characteristic indexes and moisture content was better than the index relationship. The linear models were optimum models for predicting moisture content of loess by using characteristic band (C1980) and absorption depth (D1980 and D1480) as independent variables. The calibration and validation coefficient of determination R2 and residual prediction deviation (RPD) were higher than 0.92 and 2.5, respectively, and the root mean square error (RMSE) was less than 2.5%. These results showed that the moisture content of natural soil samples can be predicted rapidly by using spectral reflectance during the soil drying process. The study can provide a reference for real-time and rapid soil moisture content monitoring and soil moisture quantitative inversion in large area by using remote sensing technology.