41 patients who took part in a withdrawal programme from long-term treatment with diazepam in a controlled clinical trial were followed up five years later. Assessments were made of outcome derived from clinical symptomatology, formal psychiatric diagnosis, psychotropic drug use and frequency of contact with both primary care and psychiatric services. Using discriminant function analysis it was found that better outcome was associated with younger patients, fewer symptoms at time of withdrawal and, more particularly, six months later, less personality disturbance, and longer duration of diazepam use before withdrawal. The implications are discussed with particular reference to policies for withdrawing benzodiazepines.