There is little rigorous evidence to underpin clinical guidelines for palliative care. However, research in palliative care is difficult, especially with dying patients. Consent is a major issue, since staff do not wish to invite dying patients to participate in trials. We, therefore, conducted a feasibility study in two units within the North West Wales NHS Trust. We explored two novel approaches to research in palliative care -cluster randomisation and randomised consent. All patients admitted to the two units during the study were asked for permission to use their data for research. We allocated the two units, at random, to use cluster randomisation or randomised consent for three months, and then to crossover to the other design. Of 24 patients dying during cluster-randomised phases, 13 gave consent on admission to use their data and were, thus, eligible to enter the trial; however, defined eligibility criteria reduced these to six active participants. Of 29 patients dying during randomised consent phases, seven gave consent on admission to use their data; although two were eligible for randomisation, neither entered the trial. We judge that cluster randomisation is the more effective design for research with dying patients. Computer simulation, based on data from 1500 dying patients on the Welsh Integrated Care Pathway, shows that crossover cluster trials need much smaller samples than simple cluster trials. Furthermore, this study has shown that crossover cluster trials are entirely feasible. We recommend a 'definitive' trial to test the crossover design more widely.