To improve their ability to estimate the survival of terminally ill cancer patients, palliative care physicians require accurate information on prognostic factors. The objective of this study was to assess the extent to which variables such as patient characteristics and primary tumor site affect the length of survival of terminally ill cancer patients. The study population consisted of 1081 cancer patients admitted for terminal care to a 15-bed palliative care unit from 1985 to 1991. Univariate Kaplan-Meier survival analysis and multivariate Cox regression analyses were used to examine the relationship between patient characteristics at admission and survival time. The factor most strongly associated with shorter survival was poor performance status; this strong relationship was not altered by taking into account sex and primary cancer site in the multivariate analysis. For patients who were bedridden at admission, the death rate was 5.5 times higher (95% confidence interval (Cl) 3.4-9.0) than that for ambulatory patients during the first four days of stay, and it was 2.8 times higher (95% Cl 2.0-3.9) subsequently (up to 19 days). The other prognostic factors significantly but slightly associated with poorer survival in the univariate analysis were primary lung cancer, male sex, and living with a spouse. These findings indicate that performance status is the main prognostic factor for accurately estimating the survival time of terminally ill cancer patients.