This paper describes the narrative discourse abilities of a group of severely traumatically brain injured (TBI) speakers at two intervals post-injury. At initial assessment (between 3 and 6 months post-injury) a group of 26 TBI speakers were assessed using a picture description task. The performance of two control groups was also examined. The first control group was comprised of 26 non-TBI orthopaedically injured speakers and the second control group comprised 26 university students. These control groups were selected in order to examine the possibility that premorbid demographic factors can influence the discourse skills of TBI speakers. Measures of interest included content (number and type of story grammar elements present), communicative efficiency (syllables per story grammar element) and pragmatic performance (errors on a modified version of Damico's Clinical Discourse Analysis - the CDA-M). The measure which most clearly separated the performance of the TBI speakers from that of the controls on an initial assessment was the CDA-M. At follow-up, a mean of 2 years 10 months post-injury, 24 of the TBI speakers were seen for review. Results at follow-up revealed significant improvement over time and the performance of the TBI speakers could no longer be differentiated from that of the orthopaedic patients included in the initial study. The findings are discussed with respect to the implications for narrative discourse sampling and measurement, in relation to severity and chronicity of brain injury.