An accurate determination of the soil organic matter (SOM) levels present in reclaimed mine soil is necessary to evaluate the success of the ecological reclamation of mines. Visible and near-infrared diffuse spectroscopy is a fast and efficient method for collecting data for soil management during the reclamation of soil removed from mines. In this work, we used spectroscopy to characterize and estimate the SOM of lands from different reclamation years after coal mining in semi-arid grasslands of North China. Our goals were: (1) to explore the SOM characteristics and the spectra of reclaimed mine soil with different reclamation ages, and (2) to establish a reliable and accurate SOM prediction model by comparing support vector machine (SVM), partial least-squares regression (PLSR), and random forest (RF) modeling methods by determining the optimal preprocessing method and input spectral region. The results showed that spectral characteristics are useful indicators for understanding progressive SOM changes in the topsoil at different reclamation ages. RF is a more appropriate method for assessing the performance of a selected spectral region and provides more accurate results with a root mean square error (RMSE) of 4.34 g center dot kg(-1) between predicted and observed SOM values. This study provides an alternative method that uses spectroscopy to estimate reclaimed soil conditions for the environmental monitoring of mining sites.