The objective of this study was to investigate which terminally ill cancer patients receive in-patient care in hospices and other specialist palliative care in-patient units. An interview survey was made of family or others who knew about the last year of life of a random sample of people who died in 1990. Twenty district health authorities from a range of inner city, outer urban and rural settings took part. Although self-selected, districts were nationally representative in terms of social characteristics and on many indicators of health service provision and usage. Interviews were obtained for 2074 cancer deaths out of a random sample of 2915, a 71% response rate. 342 had been admitted to a total of 31 different hospices. Using logistic regression analysis five factors were found to independently predict hospice in-patient care: having pain in the last year of life, having constipation, being dependent on others for help with activities of daily living for between one and six months before death, having breast cancer, and being under the age of 85 years. A third of patients with all five factors were admitted, compared with no patients with none of these factors. It was found that symptom severity, age, dependency level and site of cancer played a role in determining hospice admission but have limited predictive value. Admission seems to be governed more by chance than by need. Further research is needed to identify which patients benefit most from in-patient care in hospices and other specialist palliative care units as the present arrangements appear to be both inequitable and insupportable.