This paper describes the design and evaluation of a computer-based communication system called 'TalksBac' with four nonfluent adults with aphasia. Despite the increased availability of computer-based augmentative and alternative communication (AAC) devices, their use with adults with aphasia is limited as few devices have been designed for this population. The TalksBac system was designed specifically for nonfluent adults with aphasia and was used by four nonfluent aphasic individuals for a period of 9 months. The TalksBac system is word-based and exploits the ability of some nonfluent individuals with aphasia to recognize familiar words and short sentences. The system consists of two programs. Personal sentences and stories are entered into the TalksBac database by use of a 'carer program'. The 'user program' assists the nonfluent aphasic user to retrieve these prestored conversational items by offering probable items based on previous use of the system. The database has a hierarchical structure, but the links to individual items adapt automatically to reflect usage by individual users over time. Four nonfluent adults with aphasia were selected to participate in the study. Each subject was assessed by use of a battery of tests to provide pre-intervention data about their comprehension, expression and communication skills. Subjects and their carers were trained to use the TalksBac system and were involved in developing personalized databases. They were supported in use of the personalized systems for an intervention period of 9 months. At the end of this period, subjects' communication skills wre reassessed by use of a battery of tests. Clients' conversational abilities with and without the TalksBac system were also compared to see if use of TalksBac did augment their conversation and allow the aphasic partner to participate more fully within conversations. This was done by analysing videotaped conversations between subjects and non-aphasic partners. Results from the formal assessments indicated that there was little change in the underlying comprehension and expressive abilities of the subjects. An analysis of videotaped conversations showed that 1 subject was unable to carry out conversations using TalksBac independently, so the data for this subject was not included in the results of the analysis of conversations. Results from the video analysis for the remaining three subjects indicated that when using the TalksBac system, the conversational abilities of two subjects improved. The conversational abilities of the other subject were not enhanced by using the system as he had developed his own nonverbal strategies which he found to be more effective. This study has shown that TalksBac has the potential to augment the communication abilities of nonfluent adults with aphasia, who have not been able to develop their own compensatory strategies. Work continues to improve the efficiency of the software and to develop techniques to facilitate the carers' ability to generate conversational information for the system.