The numerous variations in verbal fluency are characteristic of oral utterances in conversation and can involve morpho-syntactic disruptions. This study focuses on self-breaks in verbal flow, whether or not they give rise to a disfluent sequence. Following Shriberg (1995) the structure description of oral interruptions (Reparandum, Interregnum, Reparans), we noted all the self-breaks along with their morpho-syntactic effects, in the eight dilogues of the CID (Corpus of Interactional Data). Our method, focusing on the self-breaks points, describes the identification and annotation procedures. It enabled us to introduce a classification of different oral phenomena relating to self-interruption and disfluency. In many cases they are followed by acoustic markers, verbal phenomena and morpho-syntactic consequences. This study made it possible to study the relationships between the interruptions themselves and their consequences. The syntagmatic process, when interrupted, was not always disrupted from a syntactic point of view: half of these ruptures are merely suspensive, the others are disfluent. The suspensive self-breaks happen with a certain regularity and their frequency has a low variation between the speakers. Considering the disfluent interruptions, it is quite different: they have a high variation. Our hypothesis is that the frequency of the suspensive breaks remaining homogeneous seems to be an essential component of speech flow. Disfluent breaks are much less frequent but the considerable variation between speakers suggests that they are representative of utterance characteristics specific to each speaker. This hypothesis is supported by the high tendency we observed: the quicker the speech, the more the speaker produces disfluent breaks. This parameter does not influence the production of suspensive self-interruptions. All types of insertion (in the Interregnum) are present in one or other of the cases of interruption although in varying degrees. The phenomenon of resuming an utterance (rather than letting it unfinished) after a disfluent self-break seemed to be a dominant characteristic of oral utterances. Moreover, if we compare the Interregnum content, these disfluencies are two kinds of separate phenomena. On another side, the percentage of disfluencies (length of self-break ratings and disfluency phenomena) compared to the length of informative content in the oral utterances varies from one speaker to the next. This ratio is relevant since the length of time spent on non-informative utterance represents at least a third of the total speaking time.