n this paper, we investigate what causes the aggregate exit rate out of unemployment estimated in Wallonia (Belgium) to decline over duration and to vary over calendar time. For that purpose, we specify a mixed proportional hazard (MPH) model where the mixing distribution depends on seasons and business cycle at the time of entry. Given its too restrictive nature, we relax the proportionality assumption in three ways. First, we allow the baseline hazard to vary non-proportionally between a boom and a recession in order to test the ranking hypothesis. Second, the variance of the mixing distribution needs not to fluctuate proportionally to its mean over calendar time at entry. Finally, we allow for random deviations from the MPH framework by introducing random cohort-specific business cycle effects at the time of exit. We estimated our model by Minimum Chi-Squares on quarterly data of male workers entering unemployment between June 1989 and February 1994. We find that the negative duration dependence of the aggregate exit rate is largely spurious. Moreover, for prime-aged, but not for young male workers, true duration dependence varies over the cycle, a finding consistent with employers ranking candidates according to unemployment duration in their recruitment decision. Changes in the composition of workers entering unemployment explain an important part of the seasonal variation of the exit rate, but not of its variation over the business cycle.