In this paper, we present a pen-shaped input device for wearable computers which reproduce and recognize three-dimensional (3D) hand motions with no external references. The input device is equipped with inertial sensors. The inertial sensors measure accelerations and angular velocities produced by a user's handwriting motion in 2D/3D spaces. The measurements from gyroscopes are integrated once to produce the attitude of the system and are consequently used to remove the effects of the gravity and the inclination of the system. The compensated acceleration measurements are doubly integrated to yield the position of the system. Due to the integration processes involved in reproducing the handwriting trajectory, the accuracy of the position measurement significantly deteriorates along with time. To reproduce handwriting trajectory of a stroke on a surface and even in the air, we propose algorithms to segment a stroke, to compensate the integration errors, and to reconstruct 2D trajectory on the x-y plane from 3D trajectory in the air. Real experimental results are included to show the effectiveness and feasibility of the proposed algorithms. The proposed methods provide a new approach for implementing a small, user-friendly keyboard alternative input device for wearable computers.