OBJECTIVE The movement for advance planning of end-of-life care was motivated in part by the assumption that medical intervention for terminally ill patients varies from what these patients would prefer. We examined the validity of this assumption by comparing actual life-sustaining treatment practices for patients in critical illness scenarios and surveyed patients' advance care preferences. RESULTS We selected at random and reviewed 7,400 inpatient medical records from a single urban teaching hospital during the period just prior to the Patient Self-Determination Act. Records of 198 patients with conditions that matched advance directive scenarios were examined, and practices to withhold or withdraw seven life-sustaining treatments were documented. Practices were compared with surveyed preferences of 102 members of the general public and 495 outpatients who were followed by the same physicians as the 198 patients. Concordance of practices and preferences for the 19 surveyed outpatients who eventually fell into one of the scenarios was also evaluated. One hundred sixty-seven inpatient cases met review criteria for the scenario coma with a small chance of recovery. Hospital patients received medical interventions that were not consistently greater or less than the preferences of the surveyed outpatients or members of the general public. Resuscitation, the most frequently withheld treatment (94% of cases), was withheld more often than surveyed preferences to decline it (56% of outpatients, p < .001). Four treatments--mechanical breathing, artificial nutrition, major surgery, and hemodialysis--were utilized comparably to surveyed outpatients' preferences (range p = .704-.055). Antibiotics and artificial hydration were withheld (9% and 6%, respectively) less often than surveyed outpatient's prior preferences to decline them (48% and 52%, respectively, p < .001 for each). Conversely, treatments given to the 19 surveyed patients who subsequently developed one of the illness scenarios were often incongruent with the patients' prior preferences. Again, in some cases more interventions were provided (26 of 63 declined treatments were given), and in some cases less (10 of 21 desired treatments were withheld). CONCLUSIONS This study does not support the assumption that, collectively, patients' advance care preferences are less interventionist than actual practices for patients in corresponding scenarios. Nevertheless, these results do support the assumption that life-sustaining treatment decisions do not conform well to individual patients' specific preferences. Progress in end-of-life care should focus on shared decision making at the patient-proxy-physician level rather than on overall life-sustaining treatments utilization.