Background: Diabetes prevalence is already high in middle income countries, particularly among older people. Current evidence on diabetes as a risk factor for dementia is limited to cohort studies in high income countries. Few studies carried out fasting glucose assessments to identify undiagnosed cases, and assess diabetes control. We aimed to determine the association between both diagnosed diabetes and total diabetes (including undiagnosed cases) and incident dementia, examining also the impact of glycaemic control on dementia risk. Methods: Population-based cohort studies of those aged 65 years and over in sites in Cuba, Dominican Republic, Puerto Rico, Peru, Venezuela, and Mexico. Diagnosed diabetes was assessed through self-reported diagnosis, and undiagnosed diabetes through fasting blood samples (glucose >= 7mmol/L). Blood pressure, smoking, underactivity and waist circumference were assessed from questionnaires and physical examination. Incident 10/66 dementia (and subtypes), and mortality, were ascertained three to five years later. Results: 12,297 interviews were completed at baseline, with 80-95% responding by site. The at risk cohort comprised 10,945 dementia-free individuals, of whom 8,171 (75%) provided blood samples. Mean age varied from 72.0 to 75.1 years by site. Total diabetes prevalence was 43.5% in Puerto Rico, ranging from 11.5% to 27.0% in other sites. Most diabetes cases (50.2% to 68.4%) were not controlled (fasting glucose > 7.0 mmol/L). 7,000 participants were followed up for 26,423 person-years with 659 incident dementia cases, and 905 dementia free deaths. Total diabetes was associated with incident 10/66 dementia (pooled meta-analysed adjusted sub-hazard ratio [pASHR] 1.25, 95% CI, 1.05-1.49, I2=48.6%), with a stronger association for uncontrolled (pASHR 1.47, 95% CI 1.19-1.81, I2=49.6%) than controlled cases (pASHR 1.29, 95% CI 0.95-1.74, I2=13.3%). Total diabetes was strongly associated with the incidence of vascular dementia (pASHR 2.25, 95% CI 1.24-4.08, I2=23.7%), but not Alzheimers Disease (pASHR 0.99, 95% CI 0.70-1.42, I2=49.0%). Conclusions: Diabetes, particularly when poorly controlled, may increase dementia risk. There is considerable scope for improved detection and control of diabetes among older people in these settings, and hence an opportunity to carry out proof of concept prevention trials. Overlapping epidemics of these age dependent disorders will challenge poorly-resourced health systems in the future.