Though 83% of 168 cancer patients admitted for “terminal care” died within 12 weeks of admission predictions of the probable length of survival showed little relation to the actual length of survival. A total of 83% of all “errors” were in an “optimistic” direction, the patient being expected to survive longer than he actually did. No significant differences were found between the accuracy of predictions made at referral by general practitioners, by doctors at other hospitals, by hospice physicians on the day of admission, or by ward sisters and senior nurses at the same time. A week after admission predicted and actual survival correlated more closely but predictions were still optimistic. It is concluded that predictions of the length of time which a cancer patient who is at the end of active treatment can be expected to survive should be made and interpreted with the greatest caution.