Short-time work programs were revived by the Great Recession. To understand their operating mechanisms, we first provide a model showing that short-time work may save jobs in firms hit by strong negative revenue shocks, but not in less severely-hit firms, where hours worked are reduced, without saving jobs. The cost of saving jobs is low because short-time work targets those at risk of being destroyed. Using extremely detailed data on the administration of the program covering the universe of French establishments, we devise a causal identification strategy based on the geography of the program that demonstrates that short-time work saved jobs in firms faced with large drops in their revenues during the Great Recession, in particular when highly levered, but only in these firms. The measured cost per saved job is shown to be very low relative to that of other employment policies.