PURPOSE: Elstein et al. (1978) emphasized the lack of a general problem solving ability in clinical diagnosis and the importance of problem features in determining performance. A crucial problem feature is difficulty, but it is usually defined post hoc (number of errors made) in studies of clinicians' diagnostic performance. A taxonomy of difficulty would allow us systematically to create/select diagnostic problems and formulate predictions for study. Such a taxonomy was developed for a study of diagnosis in Family Medicine – a specialty characterized by non-specific presentations and low prevalence of serious disease. METHODS: Categories of difficulty were derived from the psychology literature on diagnosis. They were refined, contextualised and exemplified using information from medico-legal databases, the clinical literature on diagnostic error, and interviews with family physicians. RESULTS: Diagnostic difficulty presupposes competing hypotheses. Additional difficulty comes from the way disease presents, its prevalence, and interaction with features of human reasoning: • the tendency to look for and recognize familiar patterns in the presented information, and to ignore or explain away information that does not fit a recognized pattern; • the base-rate heuristic and the associated biases, availability and representativeness; • the tendency to assume that there is only one cause for the presenting symptoms. The emerging taxonomy has the following structure: A. No easily discernible pattern in the initial patient presentation: 1) patient presents with non-specific symptom; 2) patient presents with multiple symptoms that do not make a pattern. B. A pattern can be discerned in the initial patient presentation, but 1) important features are inconsistent or missing; 2) there are distracting features; 3) the correct diagnosis is not the most prevalent; 4) there are coexisting diseases, independent or interacting. Categories of difficulty are not necessarily mutually exclusive and a patient presentation can contain more than one type of difficulty. Moreover, during diagnosis of a patient case, a clinician may pass from category A (not immediately discerning pattern) to B (pattern discerned). CONCLUSIONS: A taxonomy of diagnostic difficulty for Family Medicine was developed on the basis of psychological theory and empirical information on diagnostic error. The taxonomy aims to provide a basis for the systematic development of realistic scenarios for the study of medical diagnostic reasoning and help formulate predictions about the performance expected.