The approach described here allows to use the fuzzy Object Based Representation of imprecise and uncertain knowledge. This representation has a great practical interest due to the possibility to realize reasoning on classification with a fuzzy semantic network based system. For instance, the distinction between necessary, possible and user classes allows to take into account exceptions that may appear on fuzzy knowledge-base and facilitates integration of user's Objects in the base. This approach describes the theoretical aspects of the architecture of the whole experimental A.I. system we built in order to provide effective on-line assistance to users of new technological systems: the understanding of "how it works" and "how to complete tasks" from queries in quite natural languages. In our model, procedural semantic networks are used to describe the knowledge of an "ideal" expert while fuzzy sets are used both to describe the approximative and uncertain knowledge of novice users in fuzzy semantic networks which intervene to match fuzzy labels of a query with categories from our "ideal" expert.