Considering a fuzzy knowledge discovery system we have realized we describe here the main features of such systems. First, we consider possible methods to define fuzzy partitions on numerical attributes in order to replace continuous or symbolic attributes by fuzzy ones. We explain then how to generalize statistical indexes to evaluate fuzzy rules, detailing a special index, the intensity of implication and its generalization to fuzzy rules. We describe then one algorithm use to extract fuzzy rules. Since many fuzzy operators are available, we propose a method to choose one fuzzy conjunction, one fuzzy implication and one fuzzy aggregation, and we explain how this choice may be validated by comparing the results of the Generalized Modus Ponens applied on the premises of the examples to the effective conclusions in the database. To reduce the important number of fuzzy rules extracted, we consider also some methods to aggregate fuzzy rules, showing that usage of classical reduction schemes requires specific choices of fuzzy operators.