The potential quality of organic matters (dynamics of C and N transformations in soils), partly depends on their biochemical characteristics (Fibre components), which can be assessed by an extraction method adapted from Van Soest sequence (NDF, ADF, lignin). These characteristics can be used as input data in prediction models, for example the TAO (Transformation of Added Organics) model. A complete determination of the biochemical composition according to Van Soest is time-consuming (~1 week) and expensive (~250 euros / sample). NIRS is a non destructive, fast (<5min), and inexpensive technique which is potentially applicable to these determinations. The aim of the present study was to evaluate if the use of NIRS-predicted biochemical parameters instead of laboratory determinations led to differences in the outputs of TAO model. The performances of TAO-C simulations of C dynamics with two sets of input data were compared to the dynamics really measured during incubations in controlled conditions. The datasets were: 1) the measured biochemical composition of added materials whose kinetics were recorded; 2) the average biochemical composition estimated by NIRS for a set of added materials with the same nature but without a reference measure obtained in the laboratory. The NIR prediction was performed with calibration equations based on a heterogeneous dataset which included the different types of added materials considered in the present study. Unsurprisingly the simulations obtained with the first dataset were more accurate. However the results obtained with the second dataset were satisfactory. It can be valuably used for an interesting and rapid classification of the materials for which no chemical analysis is available: it should be possible to make an a priori prediction of their carbon dynamics type (rapid, slow, intermediary), which it would be very useful in practice. (Texte intégral)