Two types of theoretical account have been proposed to explain the phenomenon of category-specific impairment in tests of semantic memory: One stresses the importance of different cortical regions to the representation of living and nonliving categories, while the other emphasize the importance of statistical relationships among features of concepts belonging to these two broad semantic domains. Theories of the latter kind predict that the direction of a domain advantage will be determined in large part by the overall damage to the semantic system, and that the profiles of patients with progressive impairments of semantic memory are likely to include a point at which an advantage for one domain changes to an advantage for the other. The present series of three studies employed semantic test data from two separate cohorts of patients with probable dementia of Alzheimer's type (DAT) to look for evidence of such a crossover. In the first study, longitudinal test scores from a cohort of 58 patients were examined to confirm the presence of progressive semantic deterioration in this group. In the second study, Kaplan-Meier survival curves based on serial naming responses and plotted separately for items belonging to living and nonliving domains indicated that the representations of living concepts (as measured by naming) deteriorated at a consistently and significantly faster rate than those of nonliving concepts. A third study, carried out to look in detail at the performance of mildly affected patients, employed an additional cross-sectional cohort of 20 patients with mild DAT and utilized a graded naming assessment. This study also revealed no evidence for a crossover in the advantage of one domain over the other as a function of disease severity. Taken together with the model of anatomical progression in DAT based on the work of Braak and Braak (1991), these findings are interpreted as evidence for the importance of regional cerebral anatomy to the genesis of semantic domain effects in DAT.