In epidemiologic studies, unrecognized bias can contribute to observed results, causing them to be inaccurate. Analytic study designs, such as the case-control and cohort designs, each carry potential for specific forms of bias. The cohort design is not susceptible to many forms of bias that are experienced by case-control studies. A consistent "protective" effect of smoking on Alzheimer's disease was documented by many case-control studies. However, the potential effect of biases cannot be separated from the results. Cohort studies now show that smoking may either be unrelated to Alzheimer's disease onset or possibly generate a modest increased risk. In this review the results and comparisons of various studies and potential biases are discussed.