A meta-analysis, involving the secondary analysis of original data from 11 case-control studies of Alzheimer's disease, is presented for alcohol consumption and cigarette smoking. Five studies included in the meta-analysis of alcohol consumption. Alcohol consumption was computed in terms of average weekly intake, measured in ounces of 'pure alcohol'. This variable was categorized into tertiles to represent low, medium and high intake. Analyses showed no excess estimated risk of Alzheimer's disease for any level of alcohol intake. Smoking was analysed in three different manners: (1) lifetime prevalence of smoking (ever/never)--this included eight studies; (2) amount smoked (less than or equal to one pack per day versus more than one pack per day)--this included seven studies; and (3) pack-years--including four studies. A statistically significant inverse association between smoking and Alzheimer's disease was observed at all levels of analysis, with a trend towards decreasing risk with increasing consumption (p(trend) = 0.0003). A propensity towards a stronger inverse relation was observed among patients with a positive family history of dementia, but the difference between this group and the group with no such history was not statistically significant. Although the observed disturbance in nicotinic receptor function in Alzheimer's disease may provide an explanation for these findings, possible biases related to the selection or survival of study subjects cannot be fully ruled out at this time. Prospective, community-based studies of incident cases of Alzheimer's disease are needed to document in detail the smoking history, age of onset of disease and survival of patients and cognitively intact people by smoking status.