Microbial denitrification plays a central role in nitrous oxide (N2O)-emitting processes, which are involved in ecosystem services such as crop production and climate regulation. Field characterization of N2O-emitting processes being time-consuming due to great variability, laboratory determination of potential denitrification (upon incubation) is often used as a valuable test. Near infrared reflectance spectroscopy (NIRS) is a time- and cost-effective approach that has been reported to allow accurate determination of several soil properties. The objective of the present study was to assess the interest of NIRS for determining potential denitrification over a set of 460 topsoils sampled under crops, tree plantations, savanna or rainforest, originating from Madagascar, Congo-Brazzaville, Brazil, and French Guiana. Prediction of potential denitrification using NIRS was satisfying over the total set, especially with LOCAL calibration, which builds a model for each sample separately using its spectral neighbours in the calibration subset (R2 = 0.79 for validation). For the other sets, either textural or geographical, global calibration only was performed, involving for each set a unique prediction model built with all calibration samples. The accuracy of NIRS determination depended on the sample set, decreasing in the following order: Malagasy clayey set > total set ? Brazilian sandy loam set > coarse-textured set (Congo-Brazil) ? Guianese sandy clay loam set ? Congolese sandy set > non-clayey set (Congo-Brazil-Guiana), with cross-validation R2 ranging from 0.88 to 0.44 (external validation was not carried out for small-sized sets). Thus NIRS prediction was more accurate over the clayey homogeneous set than over the non-clayey heterogeneous set. As a result of global calibration, potential denitrification was expressed as a linear combination of absorbance at every wavelength. Wavelengths that contributed most to NIRS prediction of soil potential denitrificatio