Objectives: To estimate survival after a diagnosis of dementia in primary care, compared with people without dementia, and to determine incidence of dementia. Design: Cohort study using data from The Health Improvement Network (THIN), a primary care database. Setting: 353 general practices in the United Kingdom providing data to THIN. Participants: All adults aged 60 years or over with a first ever code for dementia from 1990 to 2007 (n=22 529); random sample of five participants without dementia for every participant with dementia matched on practice and time period (n=112 645). Main outcome measures: Median survival by age and sex; mortality rates; incidence of dementia by age, sex, and deprivation. Results: The median survival of people with dementia diagnosed at age 60-69 was 6.7 (interquartile range 3.1-10.8) years, falling to 1.9 (0.7-3.6) years for those diagnosed at age 90 or over. Adjusted mortality rates were highest in the first year after diagnosis (relative risk 3.68, 95% confidence interval 3.44 to 3.94). This dropped to 2.49 (2.29 to 2.71) in the second year. The incidence of recorded dementia remained stable over time (3-4/1000 person years at risk). The incidence was higher in women and in younger age groups (60-79 years) living in deprived areas. Conclusions: Median survival was much lower than in screened populations. These clinically relevant estimates can assist patients and carers, clinicians, and policy makers when planning support for this population. The high risk of death in the first year after diagnosis may reflect diagnoses made at times of crisis or late in the disease trajectory. Late recording of diagnoses of dementia in primary care may result in missed opportunities for potential early interventions.