While neuropsychological tests have been identified for the early prediction of Alzheimer's disease, this has not been established for prediction of all-cause dementia. This would be helpful for clinicians concerned about the risk of progression to dementia in patients who may present with a variety of medical and neurological conditions. We wanted to determine whether neuropsychological tests could accurately predict incident dementia within 10 and five years of diagnosis in a community-based sample. The Canadian Study of Health and Aging was conducted in three waves over a 10-year period (1991-2002). We studied 1472 non-demented participants who completed neuropsychological testing in 1991 and received a diagnostic assessment for dementia in 2001 (n = 284). We also studied 1231 non-demented participants who completed neuropsychological testing in 1996 and received a diagnostic assessment in 2001 (n = 634). Diagnosticians were blinded to performance on the predictive tests. Age, education, and sex were included as covariates in all regression analysis. Ten-year prediction: 2 tests, Rey Auditory Verbal Learning Test (RAVLT) short delayed verbal recall and Wechsler Adult Intelligence Test Revised (WAIS-R) Digit Symbol, were significant predictors of dementia (sensitivity = 78%, specificity = 72%, positive likelihood ratio = 2.81). Five-year prediction: 4 tests, Wechsler Memory Scale Information, RAVLT short delayed verbal recall, animal fluency, and WAIS-R Digit Symbol, significantly predicted incident dementia (sensitivity = 75%, specificity = 74%, positive likelihood ratio = 2.90). Regression models were supported with bootstrapping estimates. Neuropsychological tests can accurately predict progression to all-cause dementia within 10 years of diagnosis in a large community-based sample of non-demented participants.