Visible-near infrared (Vis-NIR) reflectance spectroscopy, which is rapid, cost-effective, in-situ, nondestructive and without hazardous chemicals, is increasingly being used for prediction and digital soil mapping of soil organic matter (SOM). This method is the inevitable demand for precision agriculture and soil remote sensing mapping. In the present study, the Vis-NIR (350~2 500 nm) diffuse reflectance spectral collected by ASD FieldSpec Pro FR spectrometer was truncated by removing the noisy edge values below 400 nm and above 2 450 nm and then was transformed into apparent absorbance spectral using log(1/R). Based on the relationship analysis between absorbance spectral, spectral indices and SOM, partial least squares regression (PLSR) model was applied to predict SOM, and finally the spatial variability of SOM was characterized by geostatistics method. The results indicated that good model was modeling from the characteristic bands (CB, R2=0.91, RPD=3.28) of correlation coefficient more than 0.5, the spectral index (SI) of normalized difference index (NDI, R2=0.90, RPD=3.08), CB integrating SI with which a correlation coefficient was more than 0.5 (R2=0.87, RPD=2.67), and total bands (TA, 400~2 450 nm, R2=0.95, RPD=4.36). While the digital mapping of SOM produced by kriging and cokriging interpolation methods implied a better prediction result, showing similar spatial distribution with the measured SOM, indicating that it is feasible and reliable to use these spectral indices to predict and map the spatial variability.