Hysperspectral inversion of soil salinity was researched in the present paper with the chosen study object of typical semiarid area in North Shaanxi Province. The studying sites were selected, the hyperspectral data were collected, and the soil samples were taken back for experiment analysis. The reflectance of soils (R), the logarithm of the reciprocal of the reflectance (Log(1/R)) and the continual removed reflectance (Rcr) were used to research the soil salinity. The correlations between the hyperspectral character and soil salinity was studied to filter the characteristics bands. Then the partial least squares regression (PLSR) was used to study the inversion model of soil salinity with Matlab program, and the precision was compared with the verifying sites. The research result showed that the root mean square error (RMSE) of the inversion with Rcr was the least (1.253<1.367<1.575), and its precision was the best; the correlation between the predicted value and the measured value was well (r2=0.761), the trend line was near y=x. In conclusion, the quantificational inversion model with the variables of Rcr establised by PLSR was well, which will improve the survey efficiency of soil salinity.