Abstract: In arid and semi -arid inland areas, due to high evapotranspiration , the minerals in the soil water accumulation in the soil surface, soil salinization becomes a serious threat to the local agricultural production, ecological stability and economic development. At present, most research on remote sensing technology to monitor the saline soil is focused on quantifying the relationship between the saline soil salt content and its associated, environmental factors or human factors and visible-near infrared, thermal infrared, or microwave remote sensing data. In this paper, we took the Ebinur Lake in the northeast of Junggar Basin Xinjiang as the study area. Thermal infrared emissivity spectra unique characteristics were used to determine the degree of soil salinization. First, we used the platform of FTIR (Fourier Transform Infrared Spectrometer) temperature and emissivity separation processing software to separate temperature and emissivity in order to obtain the original soil emissivity spectral data. Then, the spectrum smoothing iterative method was used to separate the soil emissivity and thermometers for elimination of environmental and human-caused errors when collecting spectral emissivity data in order to get the real soil emissivity spectral information. After that, we used Gaussian filter smoothing method to filter noise spectral data. In addition, we mixed the pure soil with salt to achieve five different soil salt contents: 0.1, 2.3 ,9.7 , 26.22, 49.8 g/kg soil and a pure soil with no salt addition, and analyzed the thermal infrared emissivity spectral characteristics of them. The raw spectral data from them were de-noised by square root transformation, logarithmic transformation, the first derivative, and the second derivative. The four transformations were compared in their normalized ratio, and we determined the relationship between spectral data and soil salinity. We also used stepwise multiple regression equation to establish six different forms of forecasting models. By comparing the models of analysis, the establishment of the square root transformation had the highest prediction accuracy and R2 was greater than 0.82. By use of stepwise multiple regression model for each data set, the modeling results very stable, and test sample coefficient of determination R2=0.82, the root mean square error(RMS) is 0.92. Prediction model had performed very well, between the Ebinur basin of thermal emissivity spectra of soil salinizaiton in square root transformation and salt content has exist a function form. In this study, we discussed the hyperspectral remote sensing monitoring technology that can be used to predict the soil salt in Ebinur Lake basin, it provided the technical methods for the large-scale, low-cost and real-time monitoring the soil salinity. Such variation of emissivity method would promote the development and application of hyperspectral remote sensing technology monitoring on the space-time dynamic of arid land saline soil for future regional ecological restoration.