Governments are showing an increasing interest in quantitative models that give insights into the determinants of unemployment duration. Yet, these models oftentimes do not explicitly take into account that unemployment prospects are influenced by personality characteristics that are not being fully captured by variables in administrative data. Using German survey data linked with administrative data, we show that numeracy skills are strongly related to unemployment duration, while at the same time we confirm well-established patterns documented in the literature. Low numeracy is strongly related to a longer unemployment duration of workers below median age (33) in our sample, even after including a rich set of controls. We find that unrealistic reservation wages are not the main driver, nor do results seem to be driven by locking-in effects caused by programme participation. On the other hand, the absence of a relationship between numeracy and unemployment duration for older workers might well be driven by a locking-in effect for those with high numeracy, as they tend to commit more often to intensive training programmes. Another tentative explanation, which is supported by the data, is that younger people have fewer signals to send such that their cognitive abilities may have a higher relative signalling value.