This study investigated the effects of a multicomponent exercise intervention on muscle strength, incidence of falls and functional outcomes in frail elderly patients with dementia after long-term physical restraint, followed by 24 weeks of training cessation. Eighteen frail elderly patients with mild dementia (88.1 ± 5.1 years) performed a multicomponent exercise program, which consisted of 4 weeks of walking, balance and cognitive exercises, followed by 4 weeks of resistance exercise performed twice weekly [8-12 repetitions at 20-50 % of the one-repetition maximum (1RM)], combined with walking, balance and cognitive exercises. Before and after training, as well as after 24 weeks of training cessation, strength outcomes, Barthel Index, balance, gait ability, rise from a chair ability, dual task performance, incidence of falls and Mini-Mental State Examination were assessed. After the first 4 weeks of training, there was a significant improvement only in the balance test, whereas no additional changes were observed. However, after the second part of the training, the participants required significantly less time for the time-up-and-go test (P < 0.05), and improved the isometric hand grip, hip flexion and knee extension strength, as well as the leg press 1RM (P < 0.01). A significant reduction was also observed in the incidence of falls (P < 0.01). After 24 weeks of training cessation, abrupt decreases were observed in nearly all of the physical outcomes (P < 0.05). The exercise intervention improved strength, balance and gait ability in frail elderly patients with dementia after long-term physical restraint, and these benefits were lost after training cessation.