To conduct a systematic review of the literature to evaluate the effectiveness of exercise interventions for people with moderate-to-severe dementia in community settings. Eight electronic databases (MEDLINE, Embase, CINAHL, AMED, PsycINFO, PEDro, The Cochrane Library and BNI) were searched from inception to July 2018. Snowball searching identified additional articles not identified initially. Articles were included if they: reported randomised or quasi-randomised controlled trials comparing exercise with usual care or no treatment; and involved people over 65 with moderate-to-severe dementia in community settings. Outcome measures of interest were strength, endurance, mobility, mood and quality of life. Titles and abstracts of all studies were screened by one reviewer. Two reviewers independently screened full text articles for all eligible studies, extracted data and assessed quality and risk of bias. Eight studies with 819 participants were included. Interventions were variable in terms of content, duration and frequency. There was some evidence exercise programmes may improve physical function of people with moderate-to-severe dementia, with significant effects seen for gait speed and endurance, and a trend towards improvement in strength. There was little evidence to suggest exercise programmes improve mood. Most studies were of low quality. Exercise was associated with improvements in gait speed and endurance for older people with moderate-to-severe dementia living in the community, but the quality of evidence was low. There was no conclusive evidence regarding effect on strength or mood. Findings are limited by the quality of the available evidence.