This study investigates the association of sleep duration with risk of all-cause mortality among elderly Brazilians using data from a 9-year population-based cohort study and applying a multivariable longitudinal categorical and continuous analysis using Coxs proportional hazards models. This analysis used data from the Bambui Health and Ageing Study (BHAS), conducted in Bambuı´ city (approximately 15 000 inhabitants) in southeastern Brazil. The study population comprised 1512 (86.8%) of all eligible 1742 elderly residents. In multivariable analysis, using sleep duration as categorical variable and controlling for multiple measures of sociodemographic and health status, those who slept 9 h or more per night were found to be at higher risk of mortality than those Who slept 7 h [hazard ratio (HR): 1.53; 95% confidence interval (CI): 1.12–2.09]. Excluding those whose deaths occurred within 2 years after entry, this association remained significant (HR: 1.56; 95% CI: 1.12–2.18). In analyses using sleep duration as a continuous variable, a linear correlationwas found between sleep duration andmortality in all adjusted models in the whole sample (HR: 1.08; 95% CI: 1.02–1.15) and following exclusion of those whose deaths occurred within 2 years after entry (HR: 1.13; 95% CI: 1.06–1.21). Both linear and quadratic terms were significant, reflecting a predicted relationship, with mortality predominantly increasing in association with long sleep duration but with the addition of a slight decrease in association with shorter sleep duration. In conclusion, long rather than short sleep duration was associated principally with all-causemortality in this sample. It is therefore reasonable to suggest that clinicians should be aware of the potential adverse prognosis associated with prolonged sleep.