Talking Bosten Lake lakeside oasis in Xinjiang，China as the study area，spectral transforms and construction of differential salinity index，ratio salinity index，and normalized salinity index were performed on the hyperspectral reflectance of soil samples，and the characteristic bands were selected preferentially by significance test to construct a geographically weighted regression-based soil salinity content estimation model based on the measured surface salinity data. The results showed that:1)Soil surface salt content averaged 7.535 g·kg-1，and the feature bands selected for spectral transformation modeling were concentrated in 466～482，1669～1728 and 1979～2371 nm，where the logarithmic reciprocal first-order differential (1/lgR)′ correlation was good and the absolute value of the correlation coefficient was 0.672； 2)The characteristic bands for the construction of the three salt indices were concentrated in 1700～1728，1992～2014 and 2375～2405 nm，and the determination coefficients of the established models were all greater than 0.870，while the determination coefficient of the spectral reflectance R was only 0.621；3)The model of the differential salt index which preferred eigenband was the best model，with the R2 of 0.934 and 0.915 for the modeling set and the test set，respectively，and the RMSE of 1.186 and 0.917，respectively.