Different textured soils (sandy loam, silty clay loam and clay) from Manas County, Xinjiang were researched with indoor spectral reflectance, continuum removal was used to process soil spectra curve, and correlation analysis was made about normalized spectral reflectance and water content for modeling. The results show that different textured soils have reflectance in a order that clay>silty clay loam>sandy loam; the critical points of field capacity in sandy loam, silty clay loam and clay were 20.01%, 24.10% and 30.43% respectively, and water content was inversely proportional to spectral reflectance below such critical points while proportional above the points. Within 1390-1623 nm band, the negative correlation coefficients of soil water content and normalized spectra reflectance show better negative correlation and reaching significant levels, R2 of the model established for soil water content prediction exceeded mostly 0.8, and the average relative error of the model was 10%. The model could accurately reflect the soil moisture content, Its advantages such as accuracy, non-destruction and rapidness provide a new approach to measuring soil water content.