There is an emerging need to apply adaptive robotic hands to substitute humans in dangerous, laborious, or monotonous work. The state-of-the-art robotic hands cannot fulfill this need, because they are expensive, hard to control and they consist of many vulnerable motors and sensors. It is aimed to develop simple, adaptive hands that are capable of grasping and holding a large variety of objects. To achieve these properties, the concept of underactuation (i.e. having fewer actuators than independently moving fingers) is applied. First new metrics are defined which quantify the range of object sizes that underactuated hands can grasp and hold. Furthermore, a new method is developed to dimension the main design parameters of underactuated hands, such that the fingers can envelope and stably grasp the required range of objects. The new performance metrics and design method are applied to the design and evaluation of a new robotic hand that consists of a minimum number of motors (i.e. one) and sensors (i.e. zero). The innovation of this hand is that it mechanically decides whether to hold an object in a precision grasp or a power grasp configuration. No sensors, auxiliary actuation mechanisms, motors or control are needed to convert between these two distinct grasp configurations. It is concluded that the principle of underactuation and the proposed design method are effective to achieve self-adaptive, robust and cheap hands that are capable to grasp and hold a large range of different objects.