$Objective:$ To systematically review and meta-analyse how physical activity (PA) changes from adolescence to early adulthood (13-30 years). $Data\; sources:$ Seven electronic databases were searched: Medline, Embase, PsycInfo, SCOPUS, ASSIA, SPORTdiscus and Web of Science. $Eligibility\; criteria\; for\; selecting\; studies:$ English-language, longitudinal studies (from 01/1980 to 01/2017) assessing PA ≥twice, with the mean age of ≥1 measurement in adolescence (13-19 years) and ≥1 in young adulthood (16-30 years) were included. Where possible, data were converted to moderate-to-vigorous physical activity (MVPA) min/day, and meta-analyses were conducted between weighted mean differences (WMDs) in adolescence and adulthood. Heterogeneity was explored using meta-regression. $Results:$ Of 67 included studies, 49 were eligible for meta-analysis. PA was lower during adulthood than adolescence WMD (95% CI) -5.2 (-7.3 to -3.1) min/day MVPA over mean (SD) 3.4 (2.6) years; heterogeneity was high (I(2) >99.0%), and no predictors explained this variation (all p>0.05). When we restricted analysis to studies with data for males (n=29) and females (n=30) separately, there were slightly larger declines in WMD (-6.5 (-10.6 to -2.3) and -5.5 (-8.4 to -2.6) min/day MVPA) (both I(2) >99.0%). For studies with accelerometer data (n=9), the decline was -7.4 (-11.6 to -3.1) and longer follow-up indicated more of a decline in WMD (95% CI) (-1.9 (-3.6 to -0.2) min/day MVPA), explaining 27.0% of between-study variation. Of 18 studies not eligible for meta-analysis, nine statistically tested change over time: seven showed a decline and two showed no change. $Conclusion:$ PA declines modestly between adolescence and young adulthood. More objective longitudinal PA data (eg, accelerometry) over this transition would be valuable, as would investigating how PA change is associated with contemporaneous social transitions to better inform PA promotion interventions. $Registration:$ PROSPERO ref:CRD42015030114.