Loess deposits are widespread, covering about 10 percent of the Earth's land surface. Magnetic susceptibility (MS), an important parameter in loess deposits, is a function of the concentration of magnetic minerals, which exhibit diagnostic features in their visible and near-infrared (NIR) spectra. Therefore, MS should correlate with spectral reflectance. To determine if a relationship exists between MS and reflectance spectra we analyzed NIR spectra and MS of over 550 samples from seven loess-paleosol sequences on Loess Plateau of Central China. NIR reflectance data were first converted into eight spectral bands and then a step-wise multiple linear regression was applied to a calibration dataset to establish a model to predict MS values. Test datasets indicate our model is very successful, all produced an r2 > 0.9. By applying this new method for estimating MS to hyperspectral remote sensing data we anticipate mapping soil quality and MS on the Loess Plateau.