In this paper we present the results of a study of high-level effects on fundamental frequency curves in French. To observe these we analysed four readings lasting approximately one minute of a text composed of three paragraphs. We used a modelling technique which allowed us to capture a number of features of individual Intonation Units embedded in a sentence, in a paragraph and finally in a text. In our data, we find that the highest intonation structure seems to be the paragraph and we suggest coding this structure for TTS systems which generally do not take into account levels of structure above that of the sentence. Finally, we propose an extension to the INTSINT transcription system in order to cover such paratone effects.