A simple analytical procedure using FT-NIR and multivariate techniques for the rapid determination of individual sugars in fruit juices was evaluated. Different NIR detection devices and sample preparation methods were tested by using model solutions to determine their analytical performance. Aqueous solutions of sugar mixtures (glucose, fructose, and sucrose; 0-8% w/v) were used to develop a calibration model. Direct measurements were made by transflection using a reflectance accessory, by transmittance using a 0.5-mm cell, and by reflectance using a fiberglass paper filter. FT-NIR spectral data were transformed to the second derivative. Partial least-squares regression (PLSR) was used to create calibration models that were cross-validated (leave-one-out approach). The prediction ability of the models was evaluated on fruit juices and compared with HPLC and standard enzymatic techniques. The PLSR loading spectra showed characteristic absorption bands for the different sugars. Models generated from transmittance spectra gave the best performance with standard error of prediction (SEP) <0.10% and R2 of 99.9% that accurately and precisely predicted the sugar levels in juices, whereas lower precision was obtained with models generated from reflectance spectra. FT-NIR spectroscopy allowed for the rapid ( ∼3 min analysis time), accurate and non-destructive analysis of sugars in juices and could be applied in quality control of beverages or to monitor for adulteration or contamination. © 2001 Elsevier Science Ltd. All rights reserved.