Near-infrared spectra were determined for samples that differed in their particle size but not in their chemical composition. Various transformations were applied to these spectra to try to reduce differences between them; small differences between the transformed spectra indicate that the effects of particle size have been almost eliminated. Also, using other samples, equations to predict chemical composition were established by applying principal component analysis to transformed and untransformed spectra, followed by multiple linear regression. Transformations that reduced the effect of particle size also increased the predictive value of the major principal components, although the effect of transformations on the accuracy of predictions decreased as the number of components used in the equations was increased. Of the transformations examined, that which seemed most useful was u″/log(1 +u), where u= log(1/reflectance).