It is shown in this article that the definitions of the traditional units of language, such as ‘sound of speech’, ‘phoneme’, ‘word’, ‘lexeme’ etc., are based on the data of two types: (I) on the evidence about the structure of language (“the primary model of language”) and (2) on the evidence about functions of these units in the text (“the secondary model of language”). As a result of this fact, such definitions are not objective and are tautological by the nature of things.