This research investigates the role of worker–firm matching algorithms in accounting for early job separation rates. For this purpose, we examine Korea’s temporary foreign worker program in which the government classifies firms by priority level and matches them with foreign workers on the basis of their stated preferences. Using administrative data, we examine predictors for the job separation rates and propose alternative matching methods using serial dictatorship. Our simulation results show that alternative matching methods can substantially reduce job separation rates, thereby suggesting a possible improvement of the Korean program.