This paper analyses theoretically and empirically how employment subsidies should be targeted. We contrast measures involving targeting workers with low incomes/abilities and targeting the unemployed under the criteria of "approximate welfare efficiency" (AWE). Thereby we can identify policies that (a) improve employment and welfare, (b) do not raise earnings inequality and (c) are self-financing. We construct a microfounded, dynamic model of hiring and separations and calibrate it with German data. The calibration shows that hiring vouchers can be AWE, while low-wage subsidies do not satisfy AWE. Furthermore, hiring vouchers targeted at the long-term unemployed are more effective than those targeted at low-ability workers.