HIV dementia has an annual incidence of 7% after AIDS development and eventually affects 20% of all HIV-infected persons. Accurate and early diagnosis of HIV dementia can lead to optimized therapeutic and management decisions. The purpose of this study was to design a valid instrument to identify HIV dementia. Five groups totalling 152 outpatients were evaluated; HIV-seronegative (SN) (n = 34); asymptomatic HIV-seropositive (ASX) (n = 38); AIDS, nondemented (AIDS) (n = 53); AIDS, mildly demented (Dm) (n = 39); and AIDS, severely demented (Ds) (n = 7). None had CNS opportunistic infections or delirium due to drug intoxication or systemic illness at the time of testing. Patients were evaluated with three different screening instruments: (a) the newly developed HIV Dementia Scale (HDS), (b) the Minimental State Exam (MMSE), and (c) the Grooved Pegboard (PB). Mean HDS scores (+/- SD) (maximum = 16) for each group were as follows: SN, 14.9 +/- 1.69; ASX, 14.1 +/- 1.72; AIDS, 12.8 +/- 3.17; Dm, 8.0 +/- 3.81; and Ds, 3.5 +/- 2.94. A Receiver-Operating Characteristic curve was used to derive an optimal HDS cut-off score of < or = 10 for identifying HIV dementia, with a sensitivity of 80%, specificity 91%, and positive predictive value 78%. The efficiencies of each instrument for identifying HIV dementia were as follows: HDS-84%, PB-86%, and MMSE-72%. The HDS is a reliable and quantitative scale that is superior to other widely used bedside tests such as the MMSE for identifying HIV dementia.