Rapid determination of soil water content is urgently needed for monitoring and modeling ecosystem processes and improving agricultural practices, especially in arid landscapes. Far-infrared band application in soil water measurement is still limited. Various samples were arranged to simulate complex field condition and emissivity was obtained from a Fourier transform infrared spectrometer. Four spectral forms (including raw spectra, logarithm of reciprocal spectra, first-order derivate, and second-order derivate) were employed to develop a partial least squares regression model. The results indicate that the model with first-order derivate spectral form was identified with the highest performance ( and ) at the range of 8.309 to . Judging from the contribution of the bands to each principal component, the band region from 8.27 to holds a great promise for soil water content estimation. Several channels of ASTER and MODIS correspond to the involved band domain, which show the potential of predicting and mapping soil water content on large scales. However, there are still constraints due to the differences in spectral resolution between instrument and sensors and the influence of complex factors under field conditions, which are still challenges for forthcoming studies.