Objective Previous meta-analyses have found that exercise prevents falls in older people. This study aimed to test whether this effect is still present when new trials are added, and it explores whether characteristics of the trial design, sample or intervention are associated with greater fall prevention effects. Design Update of a systematic review with random effects meta-analysis and meta-regression. Data sources Cochrane Library, CINAHL, MEDLINE, EMBASE, PubMed, PEDro and SafetyLit were searched from January 2010 to January 2016. Study eligibility criteria We included randomised controlled trials that compared fall rates in older people randomised to receive exercise as a single intervention with fall rates in those randomised to a control group. Results 99 comparisons from 88 trials with 19 478 participants were available for meta-analysis. Overall, exercise reduced the rate of falls in community-dwelling older people by 21% (pooled rate ratio 0.79, 95% CI 0.73 to 0.85, p<0.001, I 2 47%, 69 comparisons) with greater effects seen from exercise programmes that challenged balance and involved more than 3 hours/week of exercise. These variables explained 76% of the between-trial heterogeneity and in combination led to a 39% reduction in falls (incident rate ratio 0.61, 95% CI 0.53 to 0.72, p<0.001). Exercise also had a fall prevention effect in community-dwelling people with Parkinson's disease (pooled rate ratio 0.47, 95% CI 0.30 to 0.73, p=0.001, I 2 65%, 6 comparisons) or cognitive impairment (pooled rate ratio 0.55, 95% CI 0.37 to 0.83, p=0.004, I 2 21%, 3 comparisons). There was no evidence of a fall prevention effect of exercise in residential care settings or among stroke survivors or people recently discharged from hospital. Summary/conclusions Exercise as a single intervention can prevent falls in community-dwelling older people. Exercise programmes that challenge balance and are of a higher dose have larger effects. The impact of exercise as a single intervention in clinical groups and aged care facility residents requires further investigation, but promising results are evident for people with Parkinson's disease and cognitive impairment.