The function of soils to act as a source or sink for atmospheric CO2 depends on past and present land-use, management and climate. To estimate the potential of soils to sequester additional carbon, better understanding of the response of Soil Organic Carbon (SOC) in fractions with different turnover rates to changes in environment and management is necessary. Changes in the amount and distribution of SOC can be estimated by means of Soil Organic Matter (SOM) turnover models, which typically consist of 2 to 5 SOC pools. Ideally, these pools should correspond to measurable SOM fractions. We apply a fast method for estimating SOC in different soil fractions, which can be linked to SOM pools of a model (Roth-C). Our aim is then to estimate the site-specific potential of agricultural soils to sequester additional carbon by using this model with pools defined by the measurements. We use Diffuse Reflectance IR Fourier-Transformed (DRIFT) spectroscopy in combination with multivariate Partial Least Squares (PLS) regression to predict soil properties and SOC contents in fractions with different turnover rates. Mid-IR spectra are calibrated against known property values (e.g., SOC in a particular soil fraction), and statistical calibration models for the particular soil property are generated. The application of these statistical models to IR spectra of bulk soil samples with unknown values can then be used to predict these properties, and to validate results of SOM models. Analyses were carried out on 125 agricultural, carbonate-free, mineral topsoils from agricultural sites (arable land, meadows, pastures and alpine pasture) across Switzerland. Sites varied in altitude from 265 to 2400 m a.s.l. thus representing a gradient in annual average temperature from 10.6 down to -1.6 °C. We isolated two sensitive (particulate organic matter and dissolved organic carbon), two slow turning over (SOC associated to silt and clay or stabilised in aggregates) and a passive (oxidation resistant) SOM fraction by a combination of physical and chemical fractionations. Statistical models based on DRIFT-PLS were developed for SOC contents in these fractions and for other soil properties (e.g., clay content) on the basis of spectra from 100 soil samples. We validated the calibration with 10 independent samples and obtained an accurate prediction of the properties. We use archived and recent soil samples from long-term experiments to model the distribution and amount of SOC in these fractions and compare the Roth-C-modelled values with data predicted from DRIFT-PLS. Results from this comparison will be presented and discussed.