The chemical composition of pretreated sugarcane bagasse (SCB), in terms of cellulose, hemicellulose and lignin, was analyzed using a fast near-infrared spectroscopy (NIR) technique. Spectra of four types of SCB, prepared using ammonia, hydrothermal, organosolv, and sodium hydroxide pretreatments, were correlated with results of classical chemical analyses using partial least squares (PLS) regression. In a novel approach, isolation of the components used to prepare synthetic samples of SCB permitted assessment of their influence on the model. Inclusion of the synthetic samples did not improve the performance of the model, due to structural differences such as chemical bonding and physical interactions between the components. For natural pretreated samples, the PLS technique showed good predictive capacity in the ranges (%, w/w) of 47.2–89.4 (cellulose), 0.2–27.0 (hemicellulose), and 2.1–30.0 (lignin) with low root-mean-square error values of 4.1, 3.8, and 3.5, respectively, and coefficient of determination higher than 0.80, demonstrating the suitability of using different pretreated samples in the same calibration model.