Few patients with end-stage dementia are enrolled in hospice care. A palliative care approach would nonetheless seem to be appropriate in various care settings, including the acute care hospital. We conducted a randomized controlled trial of palliative care in patients with advanced dementia (Functional Assessment Staging Tool [FAST] stage 6d-7f) who were hospitalized with acute illness. Intervention patients received recommendations by a palliative care team with the goal of enhancing patient comfort; control patients received usual care without these recommendations. Among 99 patients enrolled over 3 years, groups were comparable at baseline in terms of gender, age, race, dementia stage, and advance directive status. Outcomes were similar in terms of mean number of hospitalizations, average length of stay, and mortality. Intervention patients were more likely than control patients to receive a palliative care plan (23% versus 4%; p = 0.008), usually on discharge, and more decisions were made to forgo certain medical treatments but the numbers were small. Fewer patients in the intervention group received intravenous therapy throughout the admission (66% vs. 81%, p = 0.025). Overall, additional interventions included daily phlebotomy for at least half of the admission (41%), systemic antibiotics (75%), and new feeding tubes (44%). Including tubes present at the time of randomization, a total of 69% received long-term enteral feeding. It was difficult for a palliative care research team to influence the care of advanced dementia patients in the acute hospital setting. When patients have advanced dementia, there may be unique barriers, including perceived prognostic uncertainty, difficulty assessing comfort level, and perceptions about tube feeding. There must be a reexamination of treatment approaches for this severely impaired group of patients. Further study should attempt to identify patients prior to the need for acute hospitalization so goals can be established when there is less urgency to make life and death decisions.