Background Elderly people with dementia are known to be less physically active compared with elderly, healthy people, emphasizing the need for interventions in order to maintain a high level of independence in activities of daily living. The aim was to evaluate the effect of long-term, group-based rehabilitation including physical activity on physical performance in elderly, community-dwelling people with mild to moderate dementia. Methods A quasi-experimental study of 18 elderly, community-dwelling people, diagnosed with mild to moderate dementia, participated in an ongoing rehabilitation programme based on integrated physical, cognitive and social activities. The outcome measure was physical performance: the 30-second sit-to-stand test, Guralnik balance test, 10-metre walking speed test, timed 6-metre walk test and a timed dual task walk test. The repeated measure ANOVA was used to analyse any overall differences between related means. Results No significant effect of time was found for the five outcome measures during the entire period. The variation in the estimate of most outcome scores was higher within subjects than between subjects during the period. Profile plots illustrated that three of the participants, who experienced severe cognitive deterioration, markedly declined in all physical performance tests. Conclusion The expected, progressive deterioration in physical performance was delayed in a small group of home-dwelling people with mild to moderate dementia participating in long-term, group-based rehabilitation. Long-term, group-based rehabilitation may have the overall potential to delay deterioration in activities of daily living performance in home-dwelling people with mild to moderate dementia; however, more studies with larger samples are needed to confirm the findings of this study.