Near-infrared reflectance spectroscopy (NIRS) analysis is based on the development of calibration equations that relate constituents and the infrared information spectrum. This study compares 2 chemometric tools for developing NIRS prediction models: the GLOBAL modified partial least squares and the recently calibration strategy known as LOCAL. The LOCAL procedure is designed to select, from a large database, samples with spectra resembling the sample being analyzed. A multispecies data sets of 2448 forage samples (1872 of grass forages, 315 legume forages and 261 forage trees) were used for the prediction of crude protein (CP), crude ash (CA), neutral detergent fiber (NDF), acid detergent fiber (ADF), and acid detergent lignin (ADL). The samples were randomly divided into a calibration set (n = 2204) and a validation set (n = 244). Spectra were collected using a Foss NIRSystems model 6500 scanning VIS/NIR spectrometer, each spectrum was collected in the range of 1108–2492 nm every 2 nm, the spectra were normalized by standard normal variate and detrend, transformation followed by a first order derivation (1,4,4,1; first derivate, 4nm gap, 4 points of first smoothing, 1 point of second smoothing). Calibration performance for each model was assessed by standard error of calibration, coefficient of determination of calibration, standard error of cross-validation, coefficient of determination of cross-validation, and residual predictive deviation. LOCAL calibration reduced the bias and produced a significant decrease in the standard prediction error compared with the GLOBAL calibration; CP (0.86 vs 0.99), CA (0.68 vs 0.79), NDF (1.29 vs 2.71), ADF (1.78 vs 2.02), and ADL (0.87 vs 1.23). The coefficient of determination values were improved using the LOCAL strategy exceeding 0.90 for all chemical constituents. The use of LOCAL algorithm accurately predict the composition of forages and could offer a practical way to develop a robust equation taking into account the biodiversity of Colombian forage resources.