Human hands employ characteristic patterns of actuation, or synergies, that contain much of the information required to describe an entire hand shape. In some cases, 80% or more of the total information can be described with only two scalar component values. Robotic hands, however, commonly only couple intra-finger joints, and rarely take advantage of this inter-finger coordination. In this paper, real-world data on a variety of human hand postures was collected using a data glove, and principal components analysis was used to calculate these synergies, resulting in what we call eigenpostures. A novel mechanism design is presented to combine the eigenpostures and drive a 17-degree-of-freedom 5-fingered robot hand. The hand uses only 2 DC motors to accurately recreate a wide range of hand shapes. We also present a design improvement that allows us to distinguish between highprecision and low-precision tasks, as well as greatly reduce overall error.