Abstract: Spectral reflectance of typical salinized soil collected from Bayannur Sumu, Horqin Middle Banner of Inner Mongolia Autonomous Region was measured, and spectral characteristics of soil with different salinization degrees were analyzed．Original spectral reflectance was transformed into square root, logarithm, logarithmic reciprocal, reciprocal and its first and second order differential, and prediction model of soil salt content was established by using multiple stepwise linear regression analysis method．Results showed that, spectral curves of soil with different salt content were basically the same in morphological characteristics．The higher the salt content, the greater the reflectivity．After mathematical transformation, positive and negative correlation between spectral reflectance and soil salt content was enhanced, especially after first-order and second-order differential transformation, correlation was significantly enhanced．For soil samples with salt content less than 7 g/kg（non-saline soil）, correlation between original spectral logarithmic first-order differential reflectance and salt content was the highest, and correlation coefficient was the highest at 1490 nm, which was −0.5898, while for soil samples with salt content greater than 7 g/kg（saline soil）, correlation between original spectral logarithmic first-order differential reflectance and soil salt content was the highest, correlation coefficient was the highest at 727 nm, which was −0.5591．In prediction model established by multiple stepwise linear regression, the second order differential model of original spectrum was the best for non-saline soil, and the R² was 0.7292．Logarithmic second order differential model was the best for saline soil, and the R² was 0.8718．It could be used for rapid determination of salt content in saline-alkali soil．