33 soil samples were collected from the reclaimed farmland in the coal mining areas in Yongcheng in order to estimate quickly and accurately SOM content by hyperspectra, in which 23 samples and 10 samples were used to establish and verify the hyperspectral models respectively. SOM contents and hyperspectra of the soil samples were measured, the correlations were analyzed between SOM content and the multi-pretreated hyperspectral data, correlation coefficients at P= 0.01 level were used to test to determine the sensitive bands, and models of simple linear regression, multiple stepwise regression (MSR) and partial least squares regression (PLSR) were established. The results showed that transformed soil spectral reflectance promoted the correlation coefficients, and PLSR models of the differential spectra were optimal after the spectral noise were reduced by multiple scattering correction. Due to little information is available now on hyperspectral inversion of SOM content in reclaimed farmland, the above results provide technical support for the efficient and rapid determination of SOM content in reclaimed farmland by using hyperspectral remote sensing.