The possible worlds model and its associated Kripkean semantics provide an intuitive semantics to epistemic logics, but they seem to commit us to model agents which are logically omniscient and perfect reasoners. In this article we show that this is not necessarily the case, if possible worlds are not considered as consistent descriptions of the real world. We propose to model the beliefs of an agent using analytic tableaux, and we suggest how beliefs can be analysed in a purely logic way, using (a modified version of) the classical analytic tableaux method. We also show a brief approach to a second dimension of analysis, the physical dimension, that will allow the user to perform tests in the real world and to add the results of these tests in the open tableaux of the logical analysis.