We conducted a co-twin control study among 50 elderly twin pairs with onsets of Alzheimer's disease (AD) separated by 3 or more years. Twenty-three male pairs (46%) were screened from the (U.S.) National Academy of Sciences-National Research Council Registry (NAS-NRC Registry) of World War II veteran twins; others (mostly women) had responded to advertisements or were referred from AD clinics. Twenty-six pairs (52%) were monozygous. The onset of AD was inversely associated with prior use of corticosteroids or ACTH (odds ratio [OR], 0.25; 95% confidence interval [CI], 0.06 to 0.95; p = 0.04). Similar but weaker trends were present among pairs discordant for history of arthritis or for prior daily use of nonsteroidal anti-inflammatory drugs (NSAIDs) or aspirin. The association was strongest when we combined use of steroids/ACTH or NSAIDs post hoc into a single variable of anti-inflammatory drugs (AIs) (OR, 0.24; CI, 0.07 to 0.74; p = 0.01). The inverse relation was strong in female (volunteer) twin pairs but was not present in the younger men from the NAS-NRC Registry. AIs had typically been taken for arthritis or related conditions, but a similar result was apparent after controlling statistically for the arthritis variable (OR, 0.08; CI, 0.01 to 0.69; p = 0.02). AIs have been proposed as a means of retarding the progression of AD symptoms, and these data suggest that AIs may also prevent or delay the initial onset of AD symptoms. Because of limitations in the case-control method, our results require corroboration with hypothesis-driven research designed to control bias and confounding.