The objective of this study was to determine whether there is an association of mortality with the type of household in elderly people. A cohort of 1, 352 elderly people aged 65 years and over at baseline in October 1992 was followed for 42 months. Follow-up was completed for 1, 266 (93.6%) (172 deceased and 1, 094 alive). From the analysis using the Kaplan-Meier method and the log-rank test, male sex, older age group (75 years and over), no satisfaction with present dwelling, disability, no use of health checks, no practices of daily preventive health promotion, no participation in social activities, and no finding life worth living (no Ikigai)were univariately statistically significantly related to mortality. Furthermore, elderly people living with their spouse only or living alone had higher survival rates than those living with their spouse and children or living with their children, and the curves among the four subclasses of household were significantly different. From the Cox proportional hazards model, living with a spouse only remained as an independent predictor for survival, and living alone was not an increased risk factor for mortality, controlling for sex, age, housing conditions, disability, use of health management, and psychosocial conditions. J Epidemiol, 1998; 8 : 65-72.