The Ebinur Lake Basin is selected as the studied area, and typical saline soil is chosen as the research object. By introducing fractional differential algorithm, setting 0.2 as the order interval and dividing 0~2 into 11 orders, the differentials of the raw spectral reflectance and its four mathematical transformations are calculated. Combining with the soil salt content measured in laboratory, the effect of fractional differential algorithm on soil hyperspectral data is explored from the perspective of correlation coefficient, standard deviation and information entropy, respectively. The results show that with the increase of the differential order, the number of bands whose correlation coefficient passes the 0.01 level of the significance test follows a decreasing trend and the 1 lg R transformation has better capacity to enhance the correlation coefficient than the other three mathematical transformations; the overall distribution of hyperspectral data becomes relatively concentrated and the difference among soil samples is gradually reduced; the total information entropy decreases, the disorder degree of the information becomes smaller and the amount of valid information increases. Fractional differential can detail the varying trend of correlation coefficient, standard deviation and information entropy, enrich the methods of hyperspectral data pre-processing, delve into spectral information based on the spectral dimension, provide a new perspective on the deep exploitation of hyperspectral data, and offer a reference for various applications of hyperspectral data, such as characteristic band selection and quantitative inversion of land surface parameters.