This communication presents the results of work on the analysis of non-directive research interviews (ENDR). The identification of a series of terms which are repeated to a significant extent at the same time in a short sequence of text (i.e. ‘bundles') enables us to see the ‘meaning packets' which make up the episodes. By following these bundles of terms, we hope to be able to locate thematic ruptures (the end of a series of bundles), i.e., the passages where the thread of the narrative unfolds as a new theme (or in this case, an episode) is developed (new series of bundles). The algorithm of the bundles has been incorporated in the semantic analysis software TROPES. We intend to demonstrate the utility of this method for the analysis of non-directive interviews. In particular we will seek to combine a lexicometric approach (bundles formed from graphic forms) and a semantic approach (bundles formed from classes of terms considered as equivalent in terms of meaning).