This paper introduces the notion of prototypicality in Ontology Engineering. Three kinds of prototypicality are considered: a concept can be more or less representative of its super-concept (conceptual prototypicality); a term can be more or less associated to a concept (lexical prototypicality); an instance can be more or less representative of its concept (instance prototypicality). Prototypicalities are modeled as order relations which allow to modulate links between concepts, terms and instances within an ontology. To calculate prototypicality gradients used to quantify these orders, we advocate a specific method which is based on all the components of an ontology (i.e. concepts, properties and instances) and a corpus. This paper also underlines the relevance of prototypicality for improving the efficiency of Ontology Engineering processes, in particular Ontology Personalization and Semantic-based Information Retrieval.