BACKGROUND: death certification data are routinely collected in most developed countries. Coded causes of death are a readily accessible source and have the potential advantage of providing complete follow-up, but with limitations. OBJECTIVE: to investigate the reliability of using death certificates for surveillance of dementia, the time trend of recording dementia on death certificates and predictive factors of recording of dementia. SUBJECTS: individuals aged 65 and over in six areas across England and Wales were randomly selected for the Medical Research Council Cognitive Function and Ageing Study (CFAS) and CFAS II with mortality follow-up. METHODS: prevalence of dementia recorded on death certificates were calculated by year. Reporting of dementia on death certificates compared with the study diagnosis of dementia, with sensitivity, specificity and Cohen's κ were estimated. Multivariable logistic regression models explored the impact of potential factors on the reporting of dementia on the death certificate. RESULTS: the overall unadjusted prevalence of dementia on death certificates rose from 5.3% to 25.9% over the last 26 years. Dementia reported on death certificates was poor with sensitivity 21.0% in earlier cohort CFAS, but it had increased to 45.2% in CFAS II. Dementia was more likely to be recorded on death certificates in individuals with severe dementia, or those living in an institution, yet less likely reported if individuals died in hospital. CONCLUSION: recording dementia on death certificate has improved significantly in the England and Wales. However, such information is still an underestimate and should be used alongside epidemiological estimations.