BACKGROUND Despite advances in modern technology of dialysis, prognosis of patients with acute renal failure (ARF) remains poor. To give the clinicians the most useful information, a model that accurately predicts outcome early in the course of ARF is required. However, because ARF is a heterogeneous syndrome and occurs in patients with diverse etiologies and some coexisting diseases, predicting outcome early is hard. The aim of this study is to evaluate prospectively the Acute Physiology and Chronic Health Evaluation (APACHE II) and organ system failure (OSF) models, evaluated prior to dialysis, in predicting hospital mortality. METHODS From June 2002 to March 2004, ARF patients requiring dialysis at Chang Gung Memorial Hospital, Chiayi, were prospectively recruited for this study. The worst clinical and laboratory data in the 24 hours before initiation of dialysis were prospectively evaluated, and the patients' APACHE II score and OSF number were assessed. RESULTS A total of 61 patients (40 male and 21 female) were enrolled, of whom 38 (62.3%) died before discharge. By multivariate logistic regression, the APACHE II score (odds ratio 1.3 per increase in one score; P<0.001), or OSF number (odds ratio 1.9 per increase in one OSF; P<0.01) and oliguria (odds ratio 4.2; P=0.04), were found to be statistically significant prognostic factors for hospital mortality. Mortality increased progressively and significantly as OSF number (chi-square for trend; P=0.001) or the APACHE II score (chi-square for trend; P < 0.001) increased. By using Youden's index, the best cut-off value for APACHE II was 24, with 63% sensitivity and 96% specificity. The best cut-off value for OSF number was 2, with a sensitivity of 81.6% and a specificity of 60.9%. The areas under the receiver operating characteristic curves for APACHE II and OSF number were 0.847 (95% confidence interval (CI)=0.752-0.942; P<0.01) and 0.769 (95% CI=0.646-892; P<0.001), respectively, indicating good model discrimination. CONCLUSIONS This study concludes that APACHE II and OSF number measured prior to initiation of dialysis reliably predict outcomes of ARF patients requiring dialysis. The mortality rates increase as the APACHE II score or OSF number increases. For predicting mortality, the APACHE II score > or = 24 was found to have 63% sensitivity and 96% specificity, and OSF number> or = 2 had 81.6% sensitivity and 60.9% specificity.