The technology of image spectroscopy has been widely used in soil attribute mapping in the past few decades.However,vegetation cover seriously affects the acquisition of soil spectral information, leading tomisestimating of soil attributes by visible and near-infrared (vis-NIR) spectroscopy. The traditional solution dealt with vegetation cover interfering soil spectra by masking out the areas with high vegetation coverage, thus resulting in absence of soil information for these areas. Some researchers also tried to use vegetation indices to estimate soil attributes, with results showing that the general applicability and transferability of these vegetation indices was limited by study areas and crop varieties. Therefore, how to remove the influence of vegetation on soil spectrum has become a crucialissue in estimating soil components, such as salt content over partially vegetated surfaces. The residual spectral unmixing method was previously used to separate different components of a mixedspectrum, however, the percentage of each component had to be known as a prerequisite. Recently blind source separation (BSS), a method previously often used in signal separation analysis, has successfully been appliedto separating soil spectral information from vegetation spectral information. In order to verify the effectiveness of BSS, in theHuanghai Raw Seed Growing Farm in Dongtai of Jiangsu, was selected as an experiment site, with its field delineated into plots diversified in soil salt content by amending the soil with salt and vegetated sparsely by seeding in different densities The experiment eventually had a total of 50 plots, 5 levels in soil salt content and 10 in sowing density. Then spectra, photos and soil samples of each plot were collected regularly until the soil surface was fully covered by vegetation. A total of 189 groups of field spectral reflectance of the plots various in vegetation coverage, soil salinity and growing season, were analyzed for influences of vegetation on estimationof soil salt content, and effectiveness of BSS removing the interference of vegetation. Results show that vegetation cover seriously affected accuracy of the estimation of soil salt content with R2_cv=0.53, RMSE_cv=3.54 g kg^-1, RPD_cv=1.47, R2_p=0.50, RMSE_p=3.33 g kg^-1 and RPD_p=1.41. However, the BSS algorithm, based on equation z=tanh(y), effectively eliminated the interference of vegetation on soil spectral reflectance, and improved accuracy of the estimation of soil salt content in the over partially vegetated areas using vis–NIRspectroscopy, with R2_cv =0.66, RMSE_cv=3.10 g kg^-1, RPD_cv=1.70, R2_p=0.63, RMSE_p=2.89 g kg^-1 and RPD_p=1.57. However, the effectiveness of BSS weakened when vegetation coverage was getting high, because it was unable to capture enough soil information from the mixed spectra. Additionally, choosing a suitable number of source spectra was essential to the results, and two was the best choice in this case. The method proposed here is expected to broaden the use of spectroscopy, which is usually limited to bare soil, and facilitates wider application of remote sensing images to map soil salinity overpartially vegetated surfaces.