A medical diagnosis support system based on a fuzzy model for illnesses is proposed. The model for each disease is a lattice structure composed by unidirectional fuzzy relations among intermediate diagnostic units (IDU). Its evaluation is sequential: it starts at the bottom units, the observable medical evidence (symptoms), and progresses through the IDUs to the cusp unit, representing the disease. Three types of IDUs are defined: for associated, non-associated and excluding evidence. These units represent medical concepts gathering knowledge under a common characteristic or criterion of diagnostic interest. Fuzzy unidirectional relations are used to quantify the medically known dependence between IDUs. The model is specified for six cardiopathies, identifying and calibrating its parameters using data from patient's records. The model performance is evaluated comparing results with the diagnosis provided by three medical observers on new patient's records. Concordance histograms are shown as an objective measure for the model performance. A discussion about application of the diagnosis sensitivity respect to symptoms and respect to IDUs relations is performed.