The hyper-spectral reflectance of soil was measured by a ASD FieldSpec within 400～1000 nm, then treated with logarithmic transformation. First derivative of soil spectra with different scales of differential window were acquired and denoised by the threshold denoising method based on wavelet transform. From the first derivative of soil spectra, feature parameters used as indicators for soil organic matter content were extracted. Results show that: (1) When the number of the scale of differential window was set as W=1~5, it is difficult to identify the spectrum contour and response feature in first derivative of soil spectra because of much noise. (2) When W=6~15, noise in first derivative of soil spectra is partly removed, and spectrum contour is identified roughly. However spectral response feature resulted from different organic content levels can not be identified clearly. (3) When W=16~30, noise in first derivative of soil spectrua is removed effectively. The coefficient of correlation between organic matter content and feature parameter MD19s is 0.803. MD19s can be used as one of the best indicators for soil organic matter content.