[en] Soil fertility diagnostics rely not only upon measurement of available nutrients but also upon the soil’s ability to retain these nutrients. Near-infrared reflectance spectroscopy (NIRS) is a rapid and non-destructive analytical technique which allows to simultaneously estimate standard soil characteristics and does not require use of chemicals. Previous studies showed that NIRS could be used in local contexts to predict soil properties. The main goal of our research is to build a methodological framework for the use of NIRS at a more global scale. The specific goals of this study were (i) to identify the best spectra treatment and processing –LOCAL versus GLOBAL regression- methods, (ii) to compare NIRS performances to standard chemical protocols and (iii) to evaluate the ability of NIRS to predict soil total organic carbon (TOC), total Nitrogen (TN), clay content and cationic exchange capacity (CEC) for a wide range of soil conditions. We scanned 1,300 samples representative of main soil types of Wallonia under crop, grassland or forest. Various sample preparations were tested prior to NIRS measurements. The most appropriate options were selected according to ANOVA analysis and multiple means comparisons of the spectra principal components. Fifteen pre-treatments were applied to a calibration set and the prediction accuracy was evaluated for GLOBAL and LOCAL modified partial least square (MPLS) regression models. The LOCAL MPLS calibrations showed very encouraging results for all the studied characteristics. On average, for crop soil samples, the prediction coefficient of variation (CVp) was close to 15% for TOC content, 7% for TN content, and 10% for clay content and CEC. The comparisons of repeatability and reproducibility of both NIRS and standard methods showed that NIRS is as reliable as reference methods. Prediction accuracy and technique repeatability allow the use of NIRS within the framework of the soil fertility evaluation and its replacement of standard protocols. LOCAL MPLS can be applied within global datasets, such as the International global soil spectral library. However, the performance of LOCAL MPLS is linked to the number of similar spectra in the dataset and more standard measurements are needed to characterize the least widespread soils.