To examine the ability of the Romberg test and the original untimed version of the Get-Up-and-Go test (GUG) to elders at risk for falls. At baseline and two annual follow-up visits, nurses administered the Romberg and GUG tests to 358 primary care patients aged 65+ years. Logistic regression models examined cross-sectional and longitudinal associations between abnormal balance tests and self-reported falls over the preceding year. Models were adjusted for age, sex, education and self-rated health (and, in the longitudinal models, for baseline falls), and the Mini-Mental State Examination (MMSE) as a measure of cognitive status. In cross-sectional analyses, falls reported at baseline were significantly associated with concurrently abnormal Romberg and GUG tests, after adjustment for covariates. In longitudinal analyses, abnormal GUG remained significantly associated with future falls, adjusting for covariates. Among those with low MMSE, GUG remained a significant predictor of future falls. Both balance tests were associated with low MMSE among those reporting no falls. Simple balance tests can help assess falls risk, particularly in cognitively impaired elderly who have elevated falls risk and might not accurately recall previous falls.