Abstract: The aim was to find out a way to measure wet expansion and dry shrinkage type soil moisture accurately and rapidly, and a method was proposed to analyze soil moisture base on near-infrared spectroscopy(NIRS). Taking yellow brown soil, Chao soil and rice soil in Hubei province as study objects, the spectra of soil samples with different soil moisture content were measured by NIR256-2.5 micro fiber spectrometer of Ocean Optic company in America in the dark room environment, corresponding soil bulk density was measured at the same time, relationship among soil water content、bulk density and spectral reflectivity was researched. Through comparative tests analysis reflecting the relationship of soil water content and spectral reflectivity by use of two soil water content representation methods and three soil spectral reflectivity representation methods, effect of soil other characters on soil reflectivity retrieval soil content was eliminated, suitable soil spectral reflectivity inversion of soil water content matching method was obtained and the surface model of such relationships and the exponent relationship model between soil volume water content of bulk density change and soil spectral reflectivity were constructed. The results showed that decision coefficients of three kinds of soil surface regression models constructed were more than 0.977, F values reached extremely significant level, and spectral reflectance and the bulk density of the partial regression coefficient were also significant or extremely significant level. The decision coefficients of exponent relationship model between soil volume water content of bulk density change and normalized reducing soil reflectivity at 1400 and 1900nm wavelength were more than 0.9, and the forecast errors were about 0.3 when the exponent model was verified and the precision was relatively good，so the built models had a good fitting effect. This study can provide a scientific reference for using near infrared spectroscopy to detect bulk density variable soil water content.