Solid biomass as a biorefinery feedstock is physically and chemically heterogeneous which stress the need for rapid and accurate methods for characterization and classification. In this study, stem wood, hemp, grape and olive residues were characterized using X-ray fluorescence (XRF) spectroscopy and principal component analysis. XRF spectra were processed by partial least squares (PLSs) regression and used to construct calibration models for several important ash-related elements. The models for the ash's phosphorous, aluminium and ash contents all had excellent predictive accuracy, with performance-to-deviation ratios above three. The models for the potassium, silica, calcium, sodium and magnesium content exhibited lower predictive accuracies. Relevant energy intervals (bins) were selected using interval PLS and variable importance on projection scores. In most cases, bin-selection increased the models' predictive accuracy and facilitated their interpretation. (c) 2012 Elsevier Ltd. All rights reserved.