Dramatic global increases in future numbers of people with dementia have been predicted. No multicentre population-based study powered to detect changes over time has reported dementia incidence. MRC Cognitive Function and Ageing Study (CFAS) undertook baseline interviews in populations aged 65+ years in England and Wales (1989-1994). Three areas (CFAS I) were selected for new sampling two decades later (2008-2011) with same geographical boundaries, sampling and approach methods (CFAS II). At 2 years CFAS I interviewed 5,156 (76% response) with 5,288 interviewed in CFAS II (74% response). Here we report a 20% drop in incidence (95% CI: 0-40%), driven by a reduction in men across all ages above 65. In the UK we estimate 209,600 new dementia cases per year. This study was uniquely designed to test for differences across geography and time. A reduction of age-specific incidence means that the numbers of people estimated to develop dementia in any year has remained relatively stable.