We tested the feasibility of a fully automated brain MRI voxel count technique--automated labeling pathway (ALP)--in a sample of 15 demented and 13 cognitively normal women (age 75-85 years) participating to the Cardiovascular Health Study (CHS). We hypothesized that ALP would replicate well-established findings of the anatomical correlates of dementia. In particular, we hypothesized that ALP volumetric measures would: (1) significantly differ between cognitively normal and demented women in those brain areas that are established markers for diagnosis of dementia (temporal and medial temporal lobes, hippocampus, amygdala and parahippocampus) but not in other brain areas (e.g., occipital lobe, visual cortex, motor cortex) and (2) correlate with visual ratings of brain disease which have been previously collected as part of the CHS. ALP required minimal operator intervention (input of brain images and verification of misalignments) and employed computer time of about 1 h per brain. ALP detected significant focal volumetric differences in the limbic system (p values between groups for hippocampus and parahippocampus: 0.002 and 0.005, respectively), temporal lobe (p < 0.0001) and caudate (p = 0.009), but not in other brain areas (e.g. occipital lobe, visual or motor cortex). Furthermore, ALP measures of medial temporal lobe atrophy strongly correlated with CHS visual ratings of ventricular enlargement (r(2) = 0.6, p = 0.002 for medial temporal lobe). In conclusion, ALP-detected focal brain atrophy was strongly associated with dementia. Because of its fully automated design, ALP technique is an ideal candidate to assess whether volumetric measures of specific areas can discriminate dementia better than currently available measures of global brain atrophy in large epidemiological studies.