We evaluated a data set consisting of 38 different materials - threats and non-threats - to determine whether there are signatures detectable in CT systems equipped with energyresolving photon-counting detectors that would help the detection task. Data with such detectors was collected in a lab setup using a Compton spectrometer and simple transmission measurements through samples of known material and thickness. Two alternative decomposition methods, SVD and NNMF, were evaluated and found to produce equivalent model predictions as a physics-based model. Furthermore, all three methods performed equally well with respect to material discrimination, which was evaluated by Kullback-Leibler divergence. No additional material signatures, beyond electron density and effective atomic number, appeared to be extractable from the attenuation spectra.