Vegetation and soil are usually both in one pixel and soil moisture content monitoring is inevitably influenced by vegetation spectrum,so it is important to eliminate the interference of vegetation spectrum.Hyperion hyperspectral data were decomposed by decomposition algorithm based on the spectrum matching to eliminate vegetation spectrum,and the first order differential and continuum removal transformation were used to dispose soil spectrum information. Then the sensitive bands were selected to establish the inversion model of soil moisture content.Results show that the best model was established by the bands X₆₆₁,X₁₀₁₉ and X₂₀₆₇ of the soil continuum-removal spectrum,and the forecasted Rz value was 0.85.When the vegetation spectrum is not eliminated,the best model was established by the bands X₅₄₁,X₉₇₉ and X₁₆₃₂ of the soil first order differential spectrum,and the forecasted Rz value was only 0.36.The method of forecasting soil water content by decomposing hyperspectral data to eliminate the vegetation spectrum is feasible and it can provide reference for the research on soil water content forecast by remote sensing.