Two methods were used to estimate premorbid IQ in a sample of 68 patients with mild dementia: (1) the National Adult Reading Test (NART), a present ability measure, and (2) an age, sex, race, education and occupation regression formula, a demographically based estimate (DIQ). The dementia sample consisted of probable Alzheimer's disease, multi-infarct dementia and a mixture of the two. The sample was divided into three levels of language disturbances (no language disturbance, naming or fluency disturbance, or both naming and fluency disturbance) based upon performance on the Visual Naming test and the Controlled Oral Word Association test. The NART IQ estimates in patients with one or more language disturbances were significantly lower than those in patients without language disturbances despite equivalent DIQ and Mini-Mental Status exam performance. The results suggest that the applicability of the NART to dementia patients with prominent language disturbances is limited.