Objectives: To examine the effects of cognitive stimulation (mahjong) and physical exercise (tai chi [TC]) on cognitive performance in persons with dementia. Design: Cluster-randomized open-label controlled design. Setting: Nursing homes. Participants: One hundred ten residents, most of whom were cholinesterase-inhibitor naive. Inclusion criteria were Mini-Mental State Examination (MMSE) = 10-24 and suffering from at least very mild dementia (Clinical Dementia Rating >= 0.5). Exclusion criteria were being bedbound, audio/visual impairment, regular activity participation before study, or contraindications for physical or group activities. Interventions: Homes were randomized into three conditions (mahjong, TC, and simple handicrafts [control]). Activities were conducted three times weekly for 12 weeks. Measurements: Primary outcome was MMSE. Secondary outcomes were immediate/delayed recall, categorical fluency, and digit span. Various biological risk factors, including apolipoprotein E epsilon 4 allele, were included as covariates. Measures were collected at 0 (baseline), 3 (posttreatment), 6, and 9 months. Results: Intent-to-treat analyses were performed using mixed-effects regression. Mahjong's effect varied by time for MMSE, delayed recall, and forward digit span. TC had similar effects but not for delayed recall. The typical pattern was that control participants deteriorated while mahjong and TC participants maintained their abilities over time, leading to enlarged treatment effects as time progressed. By 9 months, mahjong and TC differed from control by 4.5 points (95% confidence interval: 2.0-6.9; d = 0.48) and 3.7 points (95% confidence interval: 1.4-6.0; d = 0.40), respectively, on MMSE. No treatment effects were observed for immediate recall and backward digit span. Conclusions: Mahjong and TC can preserve functioning or delay decline in certain cognitive domains, even in those with significant cognitive impairment.