A typical task in chemometrics is to estimate the linear relationship between two sets of variables, i.e. the set of spectra, X, and the concentrations of some sample constituents, Y. Among the classical regression methods, partial least squares (PLS) is one of the most commonly used tools. One of the complications which could negatively affect the interpretation of the PLS model is related to the systematic variation present in X that is unrelated with the variation in Y. This situation typically occurs when X variables represent the absorbance or reflectance measured at hundreds of wavelengths, and the measurements are possibly influenced by sources of different types of variation having nothing in common with the information of interest. Orthogonal signal correction (OSC) is a recently proposed pre-processing method that seems to be promising in this context. This approach determines and removes from spectral data X the part of information which is Y-orthogonal (i.e. not correlated with Y). The purpose of the present paper is to illustrate how the technique works in application to near infrared (NIR) spectra of rapeseed meal. The results of PLS modelling for OSC pre-processed data have been compared with those of non-pre-processed as well as with those after multiplicative scatter correction (MSC). The main noticeable advantage of the OSC approach was the simplification of the calculated PLS models. It was also found that the combination of MSC with OSC may lead to improved performance of the model.