Abstract: The objective of every science lies in the isolation of units. Linguistics is no exception. The analyses of Martinet and Benveniste, dating back to the sixties, are still in everybody’s memory, even though they were largely preceded by those of Karl Bühler (1934) or Gottlob Frege (1923). The originality of this paper is to suggest that the definition of identified linguistic units should be based on semantic, and no longer formal, criteria. Besides, the kind of semantics relevant to the definition of those units is specific to each case: lexemes belong to lexical semantics, related to the theory of semantic fields and semic analysis; “connexions,” as defined by Lucien Tesnière (1959), belong to connexional semantics, based on graded relations between signata; “categories,” as defined by Jean Fourquet, belong to categorial semantics, distinguishing nominal (definiteness and number) and verbal (tense and mood) categories organized into micro-systems; “modal operators” belong to the category of judgments, separating truth judgments (concerning utterances) from reality judgments (concerning verbal processes) as well as appreciatives; and finally, particles are related to phrastic semantics with a distinction between those dealing with the organisation of utterances and others that are enunciative, especially argumentative.