Deconvolution can be a useful step in the process of modelling biological data, as it produces an overview of the information content of the data, as well as directions about the structure of the mathematical model able to describe the generating system. This paper concerns the application of a deconvolution technique, spectral analysis, to the modelling process of the concentrations of metabolites sampled in plasma during dialysis: the spectral analysis consists in linearly identifying the whole spectrum of multi-exponential decays, describing the compartmental nature of the process. The application to urea and creatinin time series provides a careful determination of the spectra of the exponential decays, thus giving interesting insight into the system kinetics: a sharp, slow decay (about 0.23 h-1 for urea and 0.17 h-1 for creatinin) affects all the subjects, whereas a variable set of smaller and faster components accounts for interpatient variability as well as for the multicompartmental nature of the process. The power ratio of the components is an index of the relative amount of volume in the related compartments. The identified spectra provide a description of the data that, although computed in a very simple way, is consistent with the results of the classical identification techniques previously applied in building compartmental models of dialysis.