This article deals with the theoretical consequences of the transcription practices. Detailed analysis of transcript excerpts taken from different authors and of their conventions is provided. It focusses on a central problem within the approaches of spoken language: the segmentation, identification, and definition of relevant units. Describing how they are visualized, spatialized, and represented, the analysis shows that very different models of interactional units and of recognizable linguistic patterns (linguistic forms categorized as belonging to a linguistic system, as being deviant or indetermined; paralinguistic phenomena holistically described or analytically articulated; forms categorized as belonging to a language or another, or categorized as hybrid) are materialized within the transcription choices. The theoretical dimension of the transcription practice is precisely located in this "embodiment" of theoretical presuppositions within the modes of representation of spoken language.