Although a relationship between marital status and mortality has long been recognized, no summary estimates of the strength of the association are available. A meta-analysis of cohort studies was conducted to produce an overall estimate of the excess mortality associated with being unmarried in aged individuals as well as to evaluate whether and to what degree the effect of marriage differs with respect to gender, geographical/cultural context, type of non-married condition and study methodological quality. All included studies were published after the year 1994, used multivariate analyses and were written in English. Pooling 53 independent comparisons, consisting of more than 250,000 elderly subjects, the overall relative risk (RR) for married versus non-married individuals (including widowed, divorced/separated and never married) was 0.88 (95% Confidence Interval: 0.85-0.91). This estimate did not vary by gender, study quality, or between Europe and North America. Compared to married individuals, the widowed had a RR of death of 1.11 (1.08-1.14), divorced/separated 1.16 (1.09-1.23), never married 1.11 (1.07-1.15). Although some evidence of publication bias was found, the overall estimate of the effect of marriage was robust to several statistical approaches and sensitivity analyses. When the overall meta-analysis was repeated with an extremely conservative approach and including eight non-significant comparisons, which were initially excluded because of data unavailable, the marriage protective influence remained significant, although the effect size was reduced (RR=0.94; 0.92-0.95). Despite some methodological and conceptual limitations, these findings might be important to support health care providers in identifying individuals "at risk" and could be integrated into the current programs of mortality risk estimation for the elderly.