I use newly digitized micro datasets to evaluate one of the first-ever labor regulations in Belgium: a maximum 9-hour workday in coal mines in 1910. On average, hourly wages and employment increased. However, these effects were size-dependent: hourly wages decreased in small firms, while all employment gains were concentrated in large firms. I rationalize these results in a directed search model where firms with heterogeneous TFP post vacancies, hours, and wages while internalizing workers' leisure preferences. Monopsony power leads to long hours. When hours are capped, a firm either increases wages to substitute lost hours with new hires, or cuts wages to restore markdowns. Which mechanism dominates depends on firm size, even holding monopsony power constant. Welfare analysis with sufficient statistics suggests that the 1910 reform improved workers' welfare.