Objective Most investigations of pharmacotherapy for treating Alzheimer's disease focus on patients with mild‐to‐moderate symptoms, with little evidence to guide clinical decisions when symptoms become severe. We examined whether continuing donepezil, or commencing memantine, is cost‐effective for community‐dwelling, moderate‐to‐severe Alzheimer's disease patients. Methods Cost‐effectiveness analysis was based on a 52‐week, multicentre, double‐blind, placebo‐controlled, factorial clinical trial. A total of 295 community‐dwelling patients with moderate/severe Alzheimer's disease, already treated with donepezil, were randomised to: (i) continue donepezil; (ii) discontinue donepezil; (iii) discontinue donepezil and start memantine; or (iv) continue donepezil and start memantine. Results Continuing donepezil for 52 weeks was more cost‐effective than discontinuation, considering cognition, activities of daily living and health‐related quality of life. Starting memantine was more cost‐effective than donepezil discontinuation. Donepezil–memantine combined is not more cost‐effective than donepezil alone. Conclusions Robust evidence is now available to inform clinical decisions and commissioning strategies so as to improve patients' lives whilst making efficient use of available resources. Clinical guidelines for treating moderate/severe Alzheimer's disease, such as those issued by NICE in England and Wales, should be revisited. © 2016 The Authors. International Journal of Geriatric Psychiatry published by John Wiley & Sons Ltd.