By matching information collected from coroner inquest records with hospital admission and discharge registers, 27 suicides and 5 other unexpected deaths were identified over 30 months among persons who either were psychiatric in-patients at the time or had been up to eight weeks previously. The relevant consultant completed a questionnaire concerning assessment and management in each case. It is suggested that such an approach might usefully be adopted widely as an audit of unexpected deaths. In 20 of the suicides the seriousness of the risk was not fully recognised; 13 absented themselves from the hospital ward without leave. Misleading clinical improvement in the absence of corresponding alleviation of situational problems, and patient alienation appeared important hazards. The findings have implications for service development, particularly when major reduction of bed numbers is planned.