Psychiatric morbidity, using the General Health Questionnaire, which detects mainly anxiety and depression, was examined in three samples of elderly people at baseline and at follow-up 2 %-3 years later. One sample comprised all people aged 85 plus who could be traced from Family Health Services Authority records living in an East End district of London (City and Hackney) and the other two samples were representative random samples of people aged 65-85, one sample from the same inner city area and the other from a semi-rural area for comparison with the younger sample (Braintree, Essex). The analyses reported here relate to changes in psychiatric morbidity at followup 2 %-3 years later. Previous analyses showed that poor health and functional ability were the strongest predictors of baseline psychiatric morbidity. The results from the follow-up data show that the most significant predictor of changes in psychiatric morbidity at follow-up was baseline psychiatric status, followed by health status. Poor health and functional ability explained IS% of the variance between the groups among the 85 plus sample and 17% and 16% in the two younger 65 plus samples. The strength of the study is its longitudinal design and verification of the results in three samples of elderly people. including one very elderly group. It contributes to the small body of literature on outcome of psychiatric morbidity. The lack of consistent associations with recovery from psychiatric morbidity (eg depression) in the literature enhances the importance of studies which aim to identify factors associated with different outcomes. WORDS-old age; elderly; housebound; functional disability; depression; psychiatric morbidity The single most common psychiatric diagnosis among random community samples of people is depression; among the elderly who live at home, depressive symptoms are disproportionately more prevalent than clinical diagnoses of depression. The proportions of elderly people who are identified as 'depressed' using the most restrictive diagnostic measures range from 0.5 to 4% (Blazer, 1989;Blazer et al., 1987), although most pre-
K E Y