Members of a terminal care support team in a London health district recorded upper and lower estimates of prognosis on 42 patients during 149 visits. Just over half of the actual survivals were within the estimate limits, and the estimates tended to be overoptimistic. For the initial observations on each patient, a health status measure, the Karnofsky index, gave a closer correlation with the actual survival than the estimates. Better predictors of prognosis are needed by staff caring for terminally ill people.