In a follow-up study of 1265 women and men aged 50, 60 and 70 years, we analysed how mortality was associated with cohabitation status (living alone/not living alone), living with/without a partner, and marital status respectively. Data originate from a longitudinal questionnaire study of a random sample of people born in 1920, 1930 and 1940 with baseline in 1990. Survival time for all individuals were established during the next 8 years until May 1998. Multivariate Cox analysis stratified by age and gender showed that individuals living alone experienced a significantly increased mortality compared to individuals living with somebody HR = 1.42(1.04-1.95) adjusted for functional ability, self-rated health, having children, smoking, diet and physical activity. Similar analyses were performed for the variable living with/without a partner HR = 1.38(1.01-1.88) and marital status HR = 1.25(0.93-1.69), adjusted for the same covariates. Inclusion of the health behaviour variables--smoking, diet and physical activity--one by one to a model with functional ability, self-rated health and one of the three determinants (cohabitation status, living with/without partner, marital status) showed no effect on the association with mortality. Hereby, we found no evidence of an indirect effect of health behaviours on the association between living arrangements and mortality. In contrast to many previous studies, we found no significant gender and age differences in the association between living arrangement and mortality. We suggest that in future studies of social relations and mortality, cohabitation status is considered to replace marital status as this variable may account for more of the variation in mortality.