Although the relation between self-rated health (SRH) and mortality is widely known, most of the studies have relied in baseline measurements unheeding the dynamics of the phenomenon. Our aim was to analyze how SRH both as a constant and as a time-dependent covariate predicts mortality in older men and women and to compare these different approaches. Subjects consisted of 110 male and 208 female (n=318) residents in the city of Jyväskylä, central Finland, aged 75 years at the baseline in 1989. The follow-up data was gathered in 1994 and mortality was followed for 10 years. Results showed that poor SRH was strongly associated with higher mortality risk in women in all models. In men, the association was found only in time-dependent and 5 year follow-up models and these associations were explained by baseline health status. To conclude, our analyses showed that there are gender differences in association between SRH and mortality and that the use of time-dependent covariate in a Cox regression model enables advantage to be taken of all the information in a longitudinal study design.