OBJECTIVES: To compare the effects of two exercise programs on proxy- and performance-based measures of activities of daily living (ADLs) and to explore potential motor and cognitive mediators underlying ADL improvements in individuals with dementia. DESIGN: A parallel, three-group, single-blind, randomized, controlled trial. SETTING: Psychogeriatric wards. PARTICIPANTS: Individuals with dementia (mean age 85.6 ± 5.1). INTERVENTIONS: Each 9-week intervention consisted of thirty-six 30-minute sessions. A combined group (n = 35) participated in two strength and two walking sessions, an aerobic group (n = 35) participated in four walking sessions, and a social control group (n = 35) participated in four nonexercise social visits per week. The social group was used as a reference for the mediation analysis, which was performed in the combined and aerobic groups. MEASUREMENTS: ADLs were assessed using the Katz index (proxy-reported ADL performance), Erlangen-ADL test (E-ADL) (instrumental ADL performance), and 7-item Physical Performance Test (PPT-7) (physical ADL performance). RESULTS: There was a group effect for Katz index (P = .02), E-ADL (P < .001), and PPT-7 (P = .003). Differences from baseline to after the intervention were similar in the combined and aerobic groups. Exercise-induced changes in global cognition mediated changes in Katz index (95% confidence interval (CI) = 0.05-0.41), leg muscle strength mediated changes in E-ADL (95% CI = 0.03-0.43), and leg muscle strength (95% CI = 0.01-0.36) and walking endurance (95% CI = 0.01-0.43) mediated changes in PPT-7 only in the combined group. CONCLUSION: Physical exercise can improve ADL levels in individuals with dementia, but improvements are small and seem independent of exercise type. Additional analyses suggest that combined aerobic and strength exercise may be more effective than aerobic-only exercise to effectively target ADL dysfunction in individuals with dementia. Future research is warranted to justify these exercise-specific findings.