Underactuated and compliant hands, frequently referred to as soft hands, have been recently proposed to overcome common issues of multifingered robotic hands. Although several prototypes have been developed, there is still a lack of systematic ways to model and control these devices to get grasps exploiting their intrinsic features. Classical tools can hardly be applied when contact surfaces are deformable and hand kinematics is not uniquely defined due to underactuation. In this letter, we propose a method to model underactuated compliant hands. The model captures how the hand closes analyzing the motion of suitable reference points defined on the hand. In particular, we present a procedure to compute the preferred grasping direction of a given hand, namely, the closure signature (CS), and then we use this information to plan power grasps. The feasibility of the proposed method has been validated by performing experiments with a soft hand fixed to a robotic arm. The use of the CS proved to increase the performance of the grasp planner. The proposed method can easily be extended to other underactuated compliant hands.