This paper presents a mechanism of coalition formation where agents solve a problem of diagnosis. Our approach considers that a diagnosis may be seen as the result of an emergent process where findings (at a microscopic level) are interpreted by entities (at a macroscopic level). At the micro-level agents interact and form coalitions. At the macro-level specialized agents are able to interpret coalition formation and infer a diagnosis. Our domain of application is student modelling and in this framework we conceive it as a diagnosis task where a ‘state of conceptions’ is ascribed to a student based on his/her problemsolving activities.