This study compared the cost-effectiveness of donepezil, a new cholinesterase inhibitor indicated for the treatment of mild-to-moderate probable Alzheimer's disease (AD), with no treatment. A Markov state transition model was employed to simulate treatment costs based on Swedish epidemiologic data. The Markov states used in the model were defined according to cognitive function, as assessed by the Mini-Mental State Examination. Data on costs and baseline transition probabilities were taken from the Kungsholmen Project, an observational, population-based study of persons aged >75 years in Sweden. Data on the treatment effect were taken from a clinical trial comparing donepezil to placebo over 24 weeks and were applied to the baseline transition probabilities to assess the effect of treating the clinical manifestations of AD in Swedish patients. Also, a within-trial analysis was performed for comparison, using transition probabilities taken from the clinical study. Both models were run for 5 years in half-year cycles, and both demonstrated various degrees of cost savings and improved effectiveness, as measured by increased time in nonsevere disease states. Thus donepezil had superior cost-effectiveness compared with no treatment.