This paper proposes methods for tracking a bare hand with a near-range depth camera attached to a video see-through Head-mounted Display (HMD) for virtual object manipulation in an Augmented Reality (AR) environment. The particular focus herein is upon using hand gestures that are frequently used in daily life. First, we use a near-range depth camera attached to HMD to segment the hand object easily, considering both skin color and depth information within arms' reaches. Then, fingertip and base positions are extracted through primitive models of the finger and palm. According to these positions, the rotation parameters of finger joints are estimated through an inverse-kinematics algorithm. Finally, the user's hands are localized from physical space by camera-tracking and then used for 3D virtual object manipulation. Our method is applicable to various AR interaction scenarios such as digital information access/control, creative CG modeling, virtual-hand-guiding, or game UIs.