SummaryThe method is based on a diagnosis data base, each diagnosis being identified by a list of signs indexed by à rough estimation of their relative frequency. Signs are terms of a tree-structured thesaurus, the relationship child – father being equivalent to implication. Furthermore, any sign is weighted according to the possibility of it being a variant characteristic rather than a pathological one. The patient’s description being provided, the credibility of each diagnosis is computed according to Glenn Shafer’s theory of belief functions, and the most plausible diagnoses are proposed for the clinician’s choice. The results obtained during a validation phase were satisfying so that we have been routinely using the model for assistance in diagnosing uncommon genetic syndromes observed in our hospital or submitted by other centers.