The soil salinization has faced a serious threat to the ecological environment in arid areas, and it is of great significance to quantitative inversion of the salt content of soil by remote sensing technology. In this paper, we gathered the farmland soil and salt crystal in Ebinur lake watershed, to prepare into soil samples with different salt content (the proportion of salt and saline soil: 0.3%~30%) in the laboratory. We measured the thermal infrared emissivity spectral of soil samples using 102F FTIR spectrometer, and through the Planck function fitting to obtain soil emissivity data, and then used the Gaussian filter method for smoothing emissivity curve to eliminate the background and noise effects. The saline soil emissivity spectral curve features were as bellow. The emissivity spectrum curve of soil with different salt content was basically consistent in shape and change tendency, and with the increase of salt content, the value of emissivity increased. Soil salinity factors had inhibitory effect on Reststrahlen absorption characteristics, which would be weakened with the increase of salt content, that presented as the depth of the asymmetric absorption valleys decreased, but the position and width changed a little. Based on the correlation analysis of emissivity and salt content, we found that: It was positively correlated between thermal infrared emissivity and salt content of soil, with the maximum correlation coefficient being 0.899, and the corresponding waveband 9.21 μm; 8.2~10.5 μm was the most sensitive wave bands for soil salinity. Using monadic linear regression, multiple stepwise regression and partial least square method to construct the prediction model, the value of R2 were respectively 0.863, 0.879 and 0.958, and RMSE were respectively 3.853%, 3.334% and 1.911%. It was proved that these three kinds of methods had certain prediction ability for salt content of soil, but partial least square was the best method. The thermal infrared wave bands of ASTER, Landsat8 and HJ-1B satellite sensors were chosen for the emissivity spectrum simulation according to the spectral response function of the sensor, and the correlation analysis results showed: ASTER’s B10, B11 and B12 bands are sensitive to the salt factor with thermal infrared spectroscopy and have a high correlation with soil salinity, and their correlation coefficient are up to 0.706, 0.786 and 0.872 respectively. Furthermore, the prediction model of soil salt content based on ASTER thermal infrared wavebands was established through the multiple linear regression method, R2 and RMSE of the predicted model was 0.833 and 3.895%. At last, the results showed that: it is feasible to quantitatively inverse salt content of saline soil by satellite thermal infrared remote sensing, which will provide a new way and reference for the remote sensing monitoring of soil salinization in arid areas.