Chromatographic signals have a specific microscopic behaviour which enables to statistically model the retention time of molecules. Such microscopic point of view is adopted in this paper for addressing the inverse problem of chro-matographic profiles inference in a Nonparametric Bayesian framework in order to propose an automatic unsupervised alternative to the traditional chemometrics methods. An application on inference on the concentration of micropollu-tants in lake water highlights the relevance of this approach when the studied mixture contains an unknown number of components.