Objectives To determine whether vision impairment is independently associated cross‐sectionally and longitudinally with dementia. Design Retrospective cohort study. Setting English Longitudinal Study of Ageing. Participants Individuals aged 50 and older Measurements Cross‐sectional association between self‐rated vision (poor or blind, moderate, normal) and dementia was analyzed, adjusting for potential confounders (sex, wealth, education, cardiovascular risk factors) using multivariable logistic regression. We also modelled the adjusted longitudinal association between vision impairment and dementia over an average of 11 years of follow‐up using Cox proportional hazards regression for individuals aged 50 to 69 and those aged 70 and older. Results After adjustment for confounders, participants who rated their vision as moderate were 2.0 (95% confidence interval (CI)=1.4–3.1) times as likely as those with normal vision to have dementia, and those who rated their vision as poor were 4.0 (95% CI=2.6–6.1) times as likely. Longitudinally, individuals aged 50 to 69 who rated their vision as moderate (1.8, 95% CI=1.0–3.0) or poor (3.6, 95% CI=1.1–11.8) were at greater risk of developing dementia than those who rated their vision as normal. There was no significant difference in risk in those aged 70 and older. Conclusion Our study confirms and extends findings from other countries, demonstrating cross‐sectional associations between moderate and poor self‐rated vision and dementia in England in all participants aged 50 and older and longitudinally over an 11‐year period in those aged 50 to 69. These results help establish vision loss as a risk factor for dementia, although it is unclear why. Research is needed to determine whether screening and treatment for vision loss may slow cognitive decline.