The ability to assess wood quality is of primary importance to the forest industry. In the case of intensive management forest such as clonal plantations of Eucalyptus where the heterogeneity of the raw material is huge, it is fundamental to be able to predict whole-tree values for wood properties from non-destructive samples. For some wood characteristics conventional methods of measurement are too time-consuming and expensive. One of the methods most likely to be adaptable to rapid measurements on wood is Near Infrared (NIR) diffuse reflectance spectroscopy. NIR spectroscopy is known for its good ability to identify compounds as it is mentioned in a lot of publications. This study evaluated the ability of NIR spectrosocopy for the assessment of some major wood characteristics within a full-sib family of a hybrid Eucalyptus. NIR diffuse reflectance spectroscopic methods is used for substitution of conventional wood analysis procedures. In this paper, the data are analysed using principal component analysis (PCA), the Mahalanobis distance and partial least squares regression (PLSR) models for measuring eight characteristics from the eucalyptus wood's NIR reflectance spectra. These characteristics are: - Extractives content, - Lignin content, - Relative proportion of lignin monomers, - Basic density, - Longitudinal growth strains, - Shrinkages in longitudinal, radial and tangential directions, - Modulus of elasticity, - Fracture energy. The first four characteristics are linked with pulp industry, the last five properties are linked with solid wood uses. On 300 samples coming from 200 trees, visible and NIR reflectance spectra are observed on a scanning spectrophotometer, and span the 400-2500 nm range. Calibration models are produced using partial least squares regression and the resulting models are validated using test sets. The root mean square error of prediction (RMSEP) is discussed. The results are critically examined. (Texte intégral)