Using data from the Survey of Income and Program Participation (SIPP) covering 1990-2011, we document that a surprisingly large number of workers return to their previous employer after a jobless spell and experience more favorable labor market outcomes than job switchers. Over 40% of all workers separating into unemployment regain employment at their previous employer; over a fifth of them are permanently separated workers who did not have any expectation of recall, unlike those on temporary layoff. Recalls are associated with much shorter unemployment duration and better wage changes. Negative duration dependence of unemployment nearly disappears once recalls are excluded. We also find that the probability of finding a new job is more procyclical and volatile than the probability of a recall. Incorporating this fact into an empirical matching function significantly alters its estimated elasticity and the time-series behavior of matching efficiency, especially during the Great Recession. We develop a canonical search-and-matching model with a recall option where new matches are mediated by a matching function, while recalls are free and triggered by both aggregate and job-specific shocks. The recall option is lost when the unemployed worker accepts a new job. A quantitative version of the model captures well our cross-sectional and cyclical facts through selection of recalled matches.