The aim of this study was to investigate the issue of whether low cognitive performance is an independent predictor of mortality in non demented elderly individuals. From an initial population sample of 2792 individuals aged 65 years and over living in Gironde (PAQUID cohort), we selected a sub-sample of 2149 individuals who had no major physical or cognitive disabilities at entry in the study. Among these 2149 individuals, 179 (8.3%) died during the three first years of follow-up. Mortality rates increased strongly with age and were significantly related with educational level, gender, smoking and presence of chronic diseases or regular use of more than 4 medications at baseline examination. Cognitive functions at entry in the study had been assessed by a battery of 6 psychometric tests. The 6 test results were analysed independently and also combined in an overall measure of cognitive functioning. Computation of mortality rates in each quartile of the 6 test scores showed a clear overall pattern of increased mortality with decreased scores. Multivariate logistic regression confirmed that subjects scoring in the lowest quartiles had the highest mortality rates, after adjustment on age, gender, educational level and other possible confounders. Odds-ratio was equal to 3.5 (95% CI: 1.5-7.9) in subjects with the lowest overall cognitive performance compared to the highest performers.