Background & Objectives : The degree and the nature of performance deficit in confrontation naming tasks of patients with dementia of the Alzheimer type (DAT) are multifaceted depending on the stage of illness. Quantative analysis of patients responses may reveal the gradual deterioration of naming ability as the illness exacerbrates. In addition, based on the cognitive model proposed by Ellis and Young (1988), functional components and network involved in naming can be scrutinized for their integrity. The aim of this study is then two fold: first, it is to determine whether thee are quantitative differences in naming performance among the normal control and three DAT groups of different severities. Secondary it is to observe whether there are qualitative differences among the groups and if so, what categories of errors could differentiate one group from the others. Method : The DAT patients were trichotomized (CDR 0.5, CDR 1, and CDR 2) according to the severity of the illness based clinical dementia rating (CDR) scale. The Korean version of Boston Naming Test (K BNT) was administered to each DAT group and the normal. The responses were analysed according to six categories and then by detailed subcategories under each of the six categories Results & Conclusion : The results revealed significant mean value differences between the normal and CDR I & 2 groups. The CDR 0.5 group differentiated from the normal group since indefinite response were more evident in the former. Moreover, semantically unrelated errors and no-response errors became predominant as the illness worsens. These findings explicate that qualitative analysis of naming errors may be a valuable tool for us to understand the nature of naming deficits in DAT patients.