We present a new framework for modelling users preferences in a fuzzy setting. Starting with a formal fuzzy context, the user enters so-called attribute dependency formulas based on his priorities. The method then yields the "interesting" formal concepts, that is, interesting from the point of view of the user. Our approach is designed for compounded attributes, i.e., attributes which include more than one trait. In this paper, after studying some properties of the formulas, we start investigating the computation of non-redundant bases for them. Such bases are wishful for a better overview of the preferences.