Background: There is need for identification of modifiable risk factors for dementia as intervention targets. Social network contact may reduce dementia risk through building cognitive reserve, but previous observational study findings are susceptible to reverse causation bias due to short follow-up. Aim: To examine the influence of social contact on incident dementia. Methods: I conducted a systematic review and meta-analysis of the association of marital status, used as a proxy measure for cumulative lifetime social contact, and dementia. I next examined the accuracy of English routinely-collected hospital data on dementia diagnosis to establish the validity of using these records to ascertain dementia status. I then used the Whitehall II prospective cohort study to examine the association of social contact frequency and incident dementia, ascertained from routinely-collected databases. Results: The pooled relative risk of dementia for people who were single or widowed, compared to married, was 1.42 (1.07, 1.90) and 1.20 (1.02, 1.41) respectively, an association which persisted after adjustment for potential confounding variables. I found that routinely collected hospital data included, during 2.5 mean years of follow-up, records of dementia for 78% of people with "gold-standard" dementia diagnosis; diagnostic recording was less likely for single people than married (odds ratio = 0.81 (0.67, 0.99)). I found that more frequent social contact at age 60 years was associated with lower risk of dementia (hazard ratio = 0.88 (0.79, 0.98)) but risk for social contact at 50 or 70 years was similar but not significantly associated. More frequent social contact during mid-life was associated with higher baseline cognition, but not subsequent rate of cognitive decline. Conclusions: More frequent social contact is likely to be associated with lower risk of subsequent dementia. This may be because social contact builds greater cognitive reserve, thereby delaying dementia onset, or that social contact is a marker of those with higher cognitive reserve.