We identify the arm of maximum expected reward subject to expected cost constraints.