Background: UK, US and European guidelines recommend the decision to initiate dialysis should be based on a combination of measurements of kidney function, nutritional status and clinical symptoms. Such recommendations assume an accurate and reproducible measure of glomerular filtration rate (GFR). Methods: Prospective study of 97 patients with chronic kidney disease (CKD) and serum creatinine >200 µmol/l (2.26 mg/dl) who between them contributed 388 24 h urine collections. Our main outcome measure was the number of patients with low residual renal function identified by different tests, using widely accepted thresholds. We calculated sensitivity, specificity, positive and negative predictive values and receiver operating characteristic curves for each comparison using a combined urea and creatinine clearance of <15 ml/min to indicate the likely presence of end stage renal disease (CKD stage 5). Results: Seventy five patients had a combined urea and creatinine clearance <15 ml/min during the study. Using the highest measurement of serum creatinine for each patient, the best of the prediction equations was the 4-variable modification of diet in renal disease (MDRD) equation (area under ROC curve 0.93). This was followed by Kt/V (AUC 0.91) and Cockroft Gault with and without correction for ideal body weight (AUC 0.89). Further analyses showed that the 4-variable MDRD equation had higher NPV (64%) but lower PPV (89%) than the other tests (NPV 40–49%, PPV 92–100%), for identifying patients whose combined clearance was <15 ml/min. Conclusion: The 4-variable MDRD formula is currently the best available prediction equation for GFR, but will nevertheless over estimate residual renal function when this is significantly impaired in up to 36% cases. Collection of 24 h urine samples may still have a role in the assessment of patients with stages 4 and 5 CKD.