Several techniques for measuring quality parameters in foods by the use of near infrared (NIR) technology have been reported. The aim of this experiment is to evaluate the main techniques in order to find the optimal measurement conditions for NIR analysis of carbohydrates in fluid food systems. Two different model systems were studied, each system containing 61 designed samples. The first system was designed to give scatter, and was based on a commercial orange juice. The other system was designed to be scatter-free, and was based on distilled water. To all samples were added the same total amounts of glucose, fructose and sucrose, and measured using the following NIR techniques: transmittance measurements in cuvettes, dry extract diffuse reflectance (DESIR), fibre-optic transflectance and fibre-optic transmittance. Calibration models were made by partial least squares regression in the spectral regions 780-2500 nm for DESIR measurements, 1100-1315 nm for 10 mm pathlengths and 1100-1880+2130-2350 nm for 1 mm pathlengths, The models were fully cross-validated. Optimal prediction errors (Root Mean Square Error of Prediction, Cross-Validated) for DESIR measurements ranged from 0.020 to 0.030% (w/w), while 1 mm cuvette values ranged from 0.008 to 0.012%, For these techniques there were only small differences between juice and water samples. Using fibre-optics, 1 mm transmittance gave values in the range 0.068-0.081% for juice samples and 0.022-0.066% for water samples, while 1 mm transflectance gave 0.044-0.051% for juice samples and 0.045-0.078% for water samples. 10 mm pathlengths provided substantially higher prediction errors than 1 mm for all techniques investigated. From these results, two main conclusions can be drawn. First, when measuring off-line, direct transmittance measurements in cuvettes gave better prediction results than DESIR, Second, when using fibre-optics, transflectance gave lower prediction errors than transmittance for scattering samples, while transmittance performed better than transflectance for non-scattering samples.
