The job finding rate declines with the duration of unemployment. While this is an old and well-established fact, it is still not well understood. Our paper makes two contributions. The first is empirical. We use “monthly search diaries”, a novel data source collected by Swiss public employment offices. A monthly search diary records all applications sent by a job seekers; and it indicates — for each single job application — whether the employer followed up with a job interview and/or a job offer. Based on more than 600,000 applications sent by 15,000 job seekers, we find negative duration dependence in applications; negative duration dependence in job interviews; but positive duration dependence in job offers (conditional on a job interview). Our second contribution is theoretical. We rationalize our empirical findings in a model of statistical discrimination, incorporating not only workers’ search decisions but also employers’ interview- and job-offer decisions. Our model captures the empirical duration patterns surprisingly well. We also provide a discussion of our evidence in light of other theories explaining duration dependence in unemployment such as: taste discrimination against the long-term unemployed, stock-flow matching, depletion of a job seeker’s personal network, and changes in application targeting/quality over time.