The main goal of the paper is to compare two punctuation systems for raw transcriptions of spoken French. Both have been designed in the perspective of natural language automatic processing. Using a case study, we show that the intuition based procedure leads, as expected, to inconsistent and not reproducible annotations, however suited to statistical processing. On the other hand, the discussion of the assumptions underlying the theoretical based system points out the shortcomings of this approach, even if it facilitates parser based processing. One of the main conclusions of the paper is that the choice of a punctuation system for oral transcriptions is strongly determined by the further practical or theoretical use planed for the transcripts.