<b><i>Background/Aims:</i></b> To examine the validity of Addenbrooke's Cognitive Examination III (ACE-III) in detecting early dementia in UK memory clinic patients aged 75-85 years. <b><i>Methods:</i></b> The ACE-III was administered to 59 patients prior to diagnosis. The extent to which scores predicted the membership of the dementia or no-dementia group was explored using receiver operating characteristic curve analysis and other parameters of diagnostic performance. Thirty-three participants (55.9%) were diagnosed with dementia (Alzheimer's disease = 56.3%, Alzheimer's disease with cerebrovascular disease = 31.3%, and vascular dementia = 12.5%). <b><i>Results:</i></b> The optimal cut-off for detecting dementia was 81/100 (scores <81 indicating dementia with a sensitivity of 0.79, a specificity of 0.96, and a positive predictive value of 0.96), with superiority over published cut-offs (88/100 and 82/100) at medium and lower prevalence rates. The number of years of full-time education had a significant positive relationship to total ACE-III scores (r = 0.697, p < 0.001) for the no-dementia group. Exploratory analysis indicated that optimal cut-offs were different for higher versus lower education groups. <b><i>Conclusions:</i></b> The ACE-III has excellent accuracy for the detection of dementia in day-to-day clinical practice. Lower cut-offs than those specified in the index paper, and the consideration of the patients' years of full-time education may be necessary for optimal diagnostic performance.