In linguistic research into Sign Languages (SL), the last two decades have been marked by the accelerated development of large corpora of discourse. One principle has practically become generalized for annotating them: that of representing gestural units by means of word-labels from the surrounding vocal language (VL) and/or from an international VL. These are termed ‘‘gloss-based notations’’ and in fact constitute the base medium for annotation in the framework of the most frequently used multimedia software in the field (e.g. ELAN). Historically speaking, resorting to the ‘‘gloss’’ procedure in VL can be easily explained by the absence of a system of transcription per se that is capable of graphically reconstructing the meaningful form of discourse in SL. The problematical nature of glosses in VLs is well known in the field. However, it is undoubtedly perceived as being more or less important depending on one’s conception of what constitutes the structural and typological specificity of SLs as opposed to VLs. In order to shed some light on the issue, we present the main lines of the theoretical perspective in which we describe French Sign Language (LSF), so as to identify just how the principle of resorting to ‘‘gloss’’ by VL words is fundamentally flawed in our opinion (Section 1). Then, in order to establish what our annotation needs would be in a concrete way, we will use a few examples to illustrate the types of difficulties we actually come up against in annotating under ELAN (Section 2).