OBJECTIVES: To identify the proportion of critically ill patients able to consent to participation in a randomised controlled trial (RCT) and to assess to what extent patient consent and relative assent processes could be conducted according to ethics committee permission. DESIGN: Descriptive study nested in an RCT. SETTING: Fifty-six UK intensive care units participating in the PAC-Man trial. PATIENTS AND PARTICIPANTS: First 500 patients consecutively enrolled into PAC-Man. MEASUREMENT AND RESULTS: The outcome measures were patient consent and/or relative assent. Of the 498 patients included, 13 (2.6%) provided consent before randomisation. Of the remaining 485 patients, relative assent was obtained for 394 patients (81.2%), and refused post-randomisation for 3 patients (0.6%). No relatives were available for 15 patients (3.1%), and it was unclear from documentation whether relative assent had been obtained for 73 patients (15.1%). Of the 482 patients who did not provide consent prior to randomisation, 188 (39%) survived. Of these, 175 (93.1%) gave retrospective informed consent, six (3.2%) refused, and seven (3.7%) did not regain mental competency. CONCLUSIONS: A very small proportion of patients were able to give consent before randomisation. Due to the high in-hospital mortality (60.6%), only around one third of the remaining patients could provide consent retrospectively. This study demonstrates difficulties experienced in obtaining consent from critically ill patients to participate in medical research and raises important issues about the ethical basis of the consent process in critical care.