It has been reported that social engagement may be associated with dementia risk. We searched PubMed, EMBASE, PsycINFO, CINAHL, LILACS, Biomed Central, Scopus, and Web of Science from January 2012 - May 2017, supplemented by extraction from previous reviews. We included cohort and case-control studies examining the association between social engagement or loneliness and dementia risk, pooling data using a random-effects model. Registered: PROSPERO (CRD42017067074). We included 31 cohort and 2 case-control studies comprising 2,370,452 participants. Poor social engagement indices were associated with increased dementia risk, including having a poor social network (RR = 1.59, 95% CI 1.31-1.96; I2 = 0.00%) and poor social support (RR = 1.28, 95% CI 1.01-1.62; I2 = 55.51%). In long-term studies (≥10 years), good social engagement was modestly protective (RR = 0.88, 95% CI 0.80-0.96; I2 = 0.00%). Loneliness was non-significantly associated with increased risk (RR = 1.38, 95% CI 0.98-1.94; I2 = 45.32). Our findings encourage interventions targeting social isolation and disengagement for dementia prevention.