Background Medical problems are often complex and ill-structured. In formulating the problem, one has to discriminate pertinent elements from irrelevant information in order to effectively find a solution. In this observation study, we describe how medical students formulate the problem of a complex case. Methods 32 third year medical students were presented with a complex case of endocarditis. They were asked to synthesize the case and give the best formulation of the problem. They were then asked to provide a diagnosis. A subsequent group of 25 students were presented with the problem already formulated and were also asked for the diagnosis. We analyzed the student's problem formulations using the presence or absence of essential elements of the case, the use of higher-order concepts and the use of relations between concepts. Results 12/32 students presented with the case made the correct diagnosis. Diagnostic accuracy was significantly associated with the use of higher-order concepts and relations between concepts. Establishing explicit relations was particularly important. Almost all students who missed the diagnosis could not elicit any relations between concepts but only reported factual observations. When presented with an already formulated problem, 19/25 students made the correct diagnosis. (p < 0.05) Conclusion When faced with a complex new case, students may not have the structured knowledge to recognize the nature of the problem. They have to build new schema or problem representation. Our observations suggest that this process involves using higher-order concepts and establishing new relations between concepts. The fact that students could recognize the disease when presented with a formulated problem but had more difficulty when presented with the original complex case indicates that knowledge of the clinical features may be necessary but not sufficient for problem formulation. Our hypothesis is that problem formulation represents a distinct ability.