Although natural languages are characterized by their great flexibility of use, linguists consider that they have a structure governed by different kinds of rules (phonological, syntactical, ). It is useful to study the concept of "structure" in the paradigm of categorial grammars notably because this notion was soon considered as problematic, for it induced a too "rigid" splitting of sentence constituents. So it becomes neceesary to change its definition and its formalization so that they can reflect the "flexible" organization of sentences. In the first part of the paper, we will show how Husserl's conceptualization of the notion of "category" governs the structure representation. Then we will analyze the structure representation in Ajdukiewicz's first categorial grammar (1935) in order to underline that this synyax presupposes that sentence structure is given or directly apparent. Therefore it is not necessary to define a procedure which will reveal it. In the second part, we will consider the changes introduced by Bar-Hilel in his bidirectional grammar (1953). This study will point out the problems that arise with Bar-Hillel's mechanical procedure of structure determination. In the last part, we show that, contrary to Ajdukiewicz's or Bar-Hillel's grammars which aimed to assign a single constituent structure to a non-ambiguous sentence, the Lambek syntactic calculus (1958 and 1988) proposes a more flexible conceptualization of the structure by introducing a richer inventory of combinatory operations in the form of a type-shifting calculus. Thanks to this historical and theoretical survey, we will show - from an epistemological point of view - the relativity of the criteria which justify the accurateness of structure representation and emphazise that the originality of current categorial grammars is that they have no independent phrase structure component, for the categories are fully instantiated structures.