It has been demonstrated that people reduce the duration of repeated words during spontaneous discourse. They do, this, presumably, to indicate that a word is 'old' or 'Given', and that the existing representation of the referent can therefore be used to simplify and facilitate comprehension. The experiment reported here was designed to replicate and extend a previous study by Fowler. The experiment extended Fowler's work in two ways; first, into reference, by comparing word repetitions that involved 'new' as distinct from Given information and, second, into memory processes, by comparing discourse from normal and amnesic subjects. Following Fowler, the experiment demonstrated that word duration is reduced for repeated words when the second utterance involves Given or old information provided that the repetition occurs before a topic change, a pattern that was observed for both normal and amnesic subjects. However, unlike Fowler, the experiment demonstrated that word duration is increased for the second utterance when that utterance involves New information, a pattern that was also observed for both normal and amnesic subjects. The experiment demonstrated that referential information is preserved and used during spontaneous discourse by amnesic subjects. The measurement of duration by acoustic analysis therefore offers a new approach to the analysis of memory processes in conversation, and the results suggest that Given/New marking involves records that create or maintain specific information about reference and/or Given/New status.