Five sets of near infrared reflectance calibration data were linearised by means of log 1/R or Kubelka-Munk functions and then subjected to a multiplicative scatter correction. The resulting calibration models were tested on sets of independent validation data. None of the treatments resulted in superior accuracy on all the data sets. Plots of correlation versus wavelength (correllelogram) for scatter corrected data are interpreted in terms of absorption bands due to the analyte, correction for overlapping bands or intercorrelations between the concentrations of analyte and other constituents in the samples. Although multiplicative scatter correction did not necessarily improve the accuracy of prediction, it did permit a better interpretation of the correllelograms.