A cross-sectional study among adult inpatients with non-organic psychiatric disorders, and among their key relatives, assessed their comprehension and recall of key information in consent forms. It also assessed their capacity to consent to participate in two hypothetical randomised controlled trials (RCTs) with different potential risks and burdens, using structured questionnaires and recorded interviews. Of the 24 participants (12 patient–key relative dyads), seven patients (58%) and three key relatives (25%) were clinically judged to lack the capacity to consent. Of the remaining 14 participants s, less than half the patients (2/5; 40%) or relatives (3/9; 33%) accurately recalled 50% of the key information on both trials. Among the eight participants (3 patients, 5 relatives) independently assessed on the MacArthur Competence Assessment Tool for Clinical Research, the proportions judged competent for each trial varied with the criteria for defining competence. No one fulfilled the stringent competence criteria for both trials. Routine assessments of the capacity of psychiatric research participants, and of relatives providing proxy consent, appear to be warranted. However, neither suboptimal understanding of consent forms, nor incompetence determined by the use of formal assessment tools, necessarily denote an incapacity to consent to research if detailed clinical assessments indicate otherwise. Research into incorporating participants' health literacy and clinical status in formal assessments may help determine the optimal standards for defining competence.