The goal of this paper is to examine the impact of unobserved heterogeneity when analysing the determinants of marriage dissolution. In the present analysis the parameter estimates of the explanatory variables appear to be insensitive to the omission of unobservables. The parameter estimates of the baseline hazard, however, are sensitive. When the unobserved heterogeneity is taken into consideration, the divorce risks increase steadily with duration. This supports the view that the declining hazard found in most studies of marital instability is due to a selection mechanism. Our analysis also demonstrates that the unobservables account for a considerable amount of the population variation in divorce propensity compared to the amount accounted for by the observed covariates.