OBJECTIVE To determine the incidence of HIV dementia and opportunistic brain disease in AIDS relative to the use of licensed antiretoviral medication (zidovudine, zalcitabine, didanosine, and stavudine). METHODS Medical records were evaluated retrospectively in a longitudinal cohort of 1109 patients with AIDS during the period 1991-4. Treatment groups were defined by start and duration of zidovudine treatment, the drugs used most often during this period were: (a) no zidovudine, (b) zidovudine before AIDS, (c) zidovudine before and after AIDS, and (d) zidovudine used in AIDS. Main outcome measures were cumulative incidence and survival from AIDS to onset of HIV dementia, progressive multifocal leukoencephalopathy (PML), cerebral toxoplasmosis, and primary CNS lymphoma. RESULTS Risk of brain disease including HIV dementia and opportunistic brain disease was reduced in patients who started zidovudine before AIDS and continued in AIDS (relative risk (RR) 0.55, 95% confidence interval (95% CI) 0.36-0.84) as well as zidovudine initiated in AIDS (RR 0.27, 95% CI 0.17-0.45) compared with untreated subjects. Treatment effects were not constant over time, decreasing by 14%-32% for each six months of follow up. This was supported by unadjusted incidences across groups stratified by duration of zidovudine use, indicating reduced risk with treatment for up to 18 months but not with longer duration of use of zidovudine. Other antiretroviral drugs had no significant effect, although these were used by only 14% of patients in this cohort. CONCLUSIONS The time limited but effective neuroprotection offered by zidovudine monotherapy for <18 months suggests that non-specific mechanisms of cerebral immunological defence may benefit from antiretroviral treatment. Due to the limitations of a retrospective study these findings require confirmation and further investigation in the context of current combination drug treatments.