Objective: To explore the effect of an activity treatment—mahjong—on the cognitive functioning of persons with mild-to-moderate dementia. Method: Participants were 62 older persons (Mage = 83.94, SD = 7.58) who met DSM-IV diagnosis of any dementia condition, had an initial Mini-Mental State Examination (MMSE) score ≤ 24, and were able to play, yet not having played mahjong for the past six months. They were randomly assigned to play either twice (n = 33) or four times (n = 29) a week over a 16-week duration. Digit forward span, digit forward sequence, verbal memory and MMSE were measured at baseline, post-test and 1-month follow-up. Results: Regardless of frequency of playing, mahjong produced consistent gains across all cognitive performance measures. It had large effect sizes on digit forward memory (1.0–1.4 for both span and sequence), moderate-to-large effect sizes on verbal memory (0.5–0.9), and a moderate effect size on MMSE (around 0.6). The effects lasted after mahjong had been withdrawn for a month, suggesting that constant practice is not necessary to achieve therapeutic effect once an initial threshold is attained. Conclusion: Mahjong is a viable treatment option for dementia. Because mahjong therapy basically does not require professional supervision and can be implemented as widely as space allows at a given time, the potential benefits of integrating mahjong into the daily routines of an institution are enormous vis-à-vis minimal, if any, cost to the institution.