The objective of the present article is to ascertain the mechanism of hyperspectral remote sensing monitoring for soil salinization, which is of great importance for improving the accuracy of hyperspectral remote sensing monitoring. Paddy soils in Wensu， Hetian and Baicheng counties of the southern Xinjiang were selected. Hyperspectral data of soils were obtained. Soil salt content (St) an electrical conductivity of 1∶5 soil-to-water extracts (EC1∶5) were determined. Relationships between St and EC1∶5 were studied. Correlations between hyperspectral indices and St, and EC1∶5 were analyzed. The inversion accuracy of St using hyperspectral technique was compared with that of EC1∶5. Results showed that: significant (p<0.01) relationships were found between St and EC1∶5 for soils in Wensu and Hetian counties, and correlation coefficients were 0.86 and 0.45, respectively; there was no significant relationship between St and EC1∶5 for soils in Baicheng county. Therefore, the correlations between St and EC1∶5 varied with studied sites. St and EC1∶5 were significantly related with spectral reflectance, first derivative reflectance and continuum-removed reflectance, respectively; but correlation coefficients between St and spectral indices were higher than those between EC1∶5 and spectral indices, which was obvious in some sensitive bands for soil salinization such as 660, 35, 1　229, 1　414, 1　721, 1　738, 1　772, 2　309 nm, and so on. Prediction equations of St and EC1∶5 were established using multivariate linear regression, principal component regression and partial least-squares regression methods, respectively. Coefficients of determination, determination coefficients of prediction, and relative analytical errors of these equations were analyzed. Coefficients of determination and relative analytical errors of equations between St and spectral indices were higher than those of equations between EC1∶5 and spectral indices. Therefore, the responses of high spectral information to St were more sensitive than those of high spectral information to EC1∶5. Accuracy of St predicted from high spectral data was higher than that of EC1∶5 estimated from high spectral data. The results of this study can provide a theoretical basis to improve hyperspectral remote sensing monitoring accuracy of soil salinization.