For instance, for the averaged spectra set, 7 samples of 23 translucent flesh mangosteen (30.4%) were correctly identified whereas 32 samples out of 92 (34.8%) were correctly classified for the individual spectra set. The higher classification accuracy may be due to each individual spectra containing more accurate translucence information than the averaged spectra. The results were partly coincident with those reported by Khuriyati et al. (2004) although the measurement mode was different. They suggested the use of 3- or 6-position spectra to obtain a highly efficient and stable calibration equation.