Infrared spectroscopy data is characterized by the presence of a huge number of variables. Applications of infrared spectroscopy in the mid-infrared (MIR) and near-infrared (NIR) bands are of widespread use in many fields. To effectively handle this type of data, suitable dimensionality reduction methods are required. In this paper, a dimensionality reduction method designed to enable effective Support Vector Machine Recursive Feature Elimination (SVM-RFE) on NIR/MIR datasets is presented. The method exploits the information content at peaks of the spectral envelope functions which characterize NIR/MIR spectra datasets. Experimental evaluation across different NIR/MIR application domains shows that the proposed method is useful for the induction of compact and accurate SVM classifiers for qualitative NIR/MIR applications involving stringent interpretability or time processing requirements.