Principal component weights are combined with regression coefficients to produce a form of spectral reconstruction which exhibits characteristics similar to the spectrum of a specific constituent. The technique is illustrated by two sets of near infrared spectra, the first consisting of wheat flour samples for which moisture and protein values were available and the second consisting of ground rapeseed meal for which oil and protein values were present. The reconstructed spectrum for water was based mainly on the second wheat flour component, which itself closely resembled the spectrum of water. The reconstructed oil spectrum matched closely the spectrum of a sample of oil extracted from the rape meal. It was not possible to extract protein from either set of samples in order to obtain a spectrum with which to compare the reconstructed spectra. However, most of the absorbance bands known to relate to proteins were present in both reconstructions, although the relative sizes of these bands differed.