We develop an optimal policy assignment rule that integrates two distinctive approaches commonly used in economics--targeting by observables and targeting through self-selection. Our method can be used with experimental or quasi-experimental data to identify who should be treated, be untreated,...