The Nirs technology is particularly well-suited to tree improvement programmes where huge numbers of samples must be analysed, but it can be used in any forestry application where the rapid measurement of wood property data is required or to screen unknown samples. Forest breeders include chemicals wood properties as selection criteria for new varieties, especially lignin, cellulose, hemicellulose. Efforts to develop multi-site, multiple-species, multiple age calibrations are rare. Because of the genotype x environment/ age interactions, we try to develop a very useful multispecies Nirs calibrations to predict chemical properties for varied eucalyptus in term of species, age, and site plantation. Global near infrared models to predict wood properties of Eucalyptus were developed using more than one hundred samples well representatives from 3,000 wood samples. These calibration set was selected on spectral data based on Mahalanobis distance. Samples were provided from different country and location (Congo, Senegal, Brazil), including different species and hybrids of Eucalyptus (E. urophylla, E. grandis, E. camaldulensis, E. urophylla x E. grandis, E. urophylla x E. pellita) from different age (from 5 to 30 years old). The global models tested by cross-validation, based on our own reference data, shown encouraged fits for extractives, lignin, cellulose, hemicelluloses. The models for lignin and extractives showed higher fits than cellulose and hemicellulose even if for these late, the model remained sufficient for prediction. The high variability of chemical properties due to the sampling, associated to the good repeatability of reference measurements, provided high values of model parameters. These results suggest that global calibration could be useful in tree breeding program and for different experiment trials from the fields, to rank genotypes for extractives, lignin, cellose, and hemicelluloses. Local models are more accurate usually and in order to get local mo