Although changes in discourse are frequently referred to in the traumatic brain injury (TBI) literature, they are difficult to objectify and measure. It is not always easy, therefore, for clinicians to differentiate between discourse behaviours which may have been present premorbidly, and those which are uniquely associated with TBI. The major aim of this study was to systematically examine and describe the nature of conversational impairment following severe TBI, with particular reference to the premorbid sociolinguistic characteristics of the TBI population. A second aim of the study was to examine the relationship between discourse impairment following TBI and severity of injury. Twenty-six TBI participants were compared with 26 non-brain-injured orthopaedic patients, and 26 university students, using Damico's Clinical Discourse Analysis (CDA). As predicted, global measures derived from the CDA did not differentiate the groups. The TBI group was, however, found to differ significantly from both control groups on a modified measure (CDA-M) which removes discourse errors that occurred with similar frequency across the three groups. Performance on this measure correlated significantly with severity of injury. Further, it was found that there were quantitative and qualitative differences between two seventy subgroups in the TBI group with respect to their CDA-M profiles. While nearly all members of the TBI group made errors associated with information transfer, only the more fundamental 'rules' of conversational interaction. The results are discussed in relation to the psychosocial implications of the findings, together with issues in sampling and measuring conversational discourse in the TBI population.