Among cultivated tropical Poaceae, sugarcane (Saccharum spp.) has the highest potential for energy production, mainly thanks to its agronomic traits. Modelling is the best way to design new sugarcane cropping systems for multi-use biomass production focusing on an energetic valuation of fibre co- and by-product. On the other hand, sugarcane industries have to rapidly adapt to changes in quality characteristics of biomass. Both require quality assessment using fast, efficient and robust analytical methods to determine biomass characteristics and/or to adjust processes. In this study, near infrared (NIR) reflectance spectroscopy was assessed for the prediction of the lignocellulosic compounds of sugarcane biomass. A total of 228 samples were taken from three genotypes grown at four contrasting locations in Reunion Island (SW Indian Ocean) and harvested at three ages during one plant crop cycle. The field samples were separated into five anatomical parts (millable stalk, top of the stalk, green leaf blade, green leaf sheath and trash), ground using two different methods and then analysed by the sequential van Soest method. Finally, 456 powders were scanned using a NIR XDS monochromator. Modified partial least square (MPLS) regression was applied on spectra scatter-corrected with standard normal variate and detrend followed by second derivative (SNVD-D2). Four calibration models were developed from leave-one-out location calibration data sets. To avoid over- optimistic results, independent validation was carried out at each location. This original validation method demonstrated the actual potential of our NIR model to predict the lignocellulosic compounds of independent sugarcane samples. At the same time, the performance of the NIR model will facilitate the timely supply of reference values for use in ecophysiological growth models.