In this paper, we investigate whether the diverging evolution of unemployment in the two main regions of Belgium, Flanders and Wallonia, is driven by different evolutions in their average unemployment duration and/or their unemployment incidence. To that purpose, we proceed in two stages. In the first stage, we estimate a mixed proportional hazard model by region, and decompose variations inthe aggregate outflow rate over calendar time between a general effect, i.e.changes in the exit rate of all currently unemployed, and a composition effect,i.e. fluctuations in the average quality of entrants. We also specify a non-proportional model to check whether the general effect is the same for unemployed workers with different schooling levels and sub-region ofliving, in each region. In the second stage, we decompose variations of the unemployment stock in Flanders and Wallonia into an incidence effect and the duration effect estimated in the first stage. We base our analysis on yearly exit probabilities of male workers aged 25-44 and flowing into unemployment in June, each year from 1972 to 1992. The use of aggregate data covering 21 years and stratified by a set of relevant characteristics allows us to control for changes in the inflow composition without relying(completely) on functional form assumptions.