Objective: To evaluate intervention and outcome in critically ill patients treated with high-volume haemofiltration (HV-HF). Design: Prospective cohort analysis. Setting. 18-bed closed format general intensive care unit (ICU) of a teaching hospital. Patients: 30-month cohort of ICU patients treated with HV-HF. Interventions: Intermittent high-volume venovenous haemofiltration. Endpoints: Observed and predicted mortality in prospectively stratified prognostic groups. Measurements and results: Clinical and filtration data, Acute Physiology and Chronic Health Evaluation (APACHE) II, Simplified Acute Physiology Score (SAPS) II and the Madrid Acute Renal Failure (ARF) score and predicted mortality. A total of 306 patients were haemofiltrated (140 medical, 166 surgical), 52% were oliguric. Mean APACHE II score was 31 (SD 8) and mean SAPS II score 60 (SD 16). Mean ultrafiltrate rate was 63 ml/min (SD 20). A median total of 160 litres (90% range 49 to 453) were filtrated per patient, material costs were 565 ECU (90% range 199 to 1514). ICU mortality was 33%, hospital mortality 40% [95% confidence interval (CI) 34 to 45], predicted mortality by the ARF score 67% (CI 66 to 69). Non-cardiac surgery mortality was 47% (CI 39 to 54), 73% (CT 70 to 76) predicted by APACHE II and 67% (CI 64 to 70) by SAPS II. Observed mortality was significantly lower than predicted in all prognostic groups. The standardised mortality ratio (SMR) was no higher than the SMR in the overall ICU population. Conclusions: Mortality in HV-HF patients was lower than that predicted by illness severity scores, as was the case in all patients in our ICU. Treatment with HV-HF appears to be safe and feasible. The efficacy of HV-HF should be tested in randomised, controlled trials of suitable power.