Empirical analyses of twin mortality often use models with dependent unobserved frailty terms capturing genetic and childhood environmental determinants. This ignores that mortality rates can be co-dependent due to bereavement effects, i.e. to a time-dependent causal effect of the loss of the co-twin on the mortality rate of the surviving twin. We develop a novel methodology and perform an empirical analysis based on a comprehensive model incorporating both types of dependence. We prove full identification without functional-form restrictions and we estimate models with data on twin pairs from the Danish Twin Registry. Among men, the loss of an identical co-twin at age 75 causally reduces the remaining lifetime on average by more than a year. This bereavement effect is less severe among non-identical twins or if the loss occurs at a higher age. Estimates of correlations between the frailty terms by zygosity and the ensuing implications for the relative importance of mortality determinants are highly sensitive to whether bereavement is taken into account.