Background and aims: Few studies have investigated the reliability of formulas estimating renal function in very old people. Methods: We studied 154 elderly people (mean age: 82 yrs). Serum creatinine (SC) was measured by the Jaffé method, and creatinine clearance (CrCl) with 24-h urine collection. Agreement was measured with the average ratio estimated/measured CrCl, and precision with the 95% agreement intervals (95% AI). We calculated the proportion of residents correctly classified as having renal insufficiency (accuracy). Results: The Cockcroft-Gault (CG) and Modification of Diet in Renal Disease 1 (MDRD1) formulas showed good average agreement with measured CrCl (0.95 and 1.016, respectively); the MDRD2 formula was more biased. Results were consistent in women, whereas the MDRD1 was more biased in men (average ratio: 1.196). The 95% AI showed that all formulas can yield results as low as 50% or as high as 200% of measured CrCl. The proportion of people with CrCl<60 ml/min misclassified by the CG, MDRD1, and MDRD2 formulas as having normal renal function was 21.4%, 27.0%, and 38.8%, respectively. These results were consistent across the various subgroups, especially in subjects with normal SC. Conclusions: The clinical usefulness of formulas commonly used to estimate CrCl was limited, regardless of subjects’ characteristics.