Rising costs of intensive care and the ability to prolong the life of critically ill patients creates a need to recognise early those patients who will die despite treatment. We used changes in a modified APACHE II score (organ failure score) to make daily predictions of individual outcome in 3600 patients. 137 patients were predicted to die and of these, 131 (95.6%) died within 90 days of discharge from hospital (sensitivity 23.4%, specificity 99.8%); a false-positive diagnosis rate of 4.4%. 2 of the 6 survivors have subsequently died but 4 are alive with good quality of life. Patients predicted to die stayed 1492 days in intensive care and incurred 16.7% of total intensive care expenditure and 46.4% of the cost of all patients that died. Median survival after a prediction to die was 2 days, accounting for 62% of intensive care patient days in this patient group, giving an effective intensive care cost per survivor of UK 129,651 pounds. If used prospectively, this algorithm has the potential to indicate the futility of continued intensive care but at the cost of 1 in 20 patients who would survive if intensive care were continued.