Arm reaching towards remembered targets in three-dimensional space was analyzed. The aim of the project was to test whether various amounts of visual feedback, in combination with constant forces applied to the arm during reaching would affect the magnitude and direction of the reaching errors. The robotic arm (Haptic Master, Moog Inc) recorded the spatial position of the pointer attached to its end effector, thus tracking the movement of the subject's arm. Three haptically rendered targets at different points in space were presented to the subjects using stereo virtual environment. The simulation was programmed using Visual C++ and OpenGL. Eight subjects were asked to remember the position of the target and then to reach the target with the pointer. Three different types of visual feedback were used, with full vision (Vision), with the target disappearing immediately before the movement onset (No Vision) and one second after the movement onset (Intermediate Vision). In all three visual conditions, the subjects arm was visible during the movement. In addition, an external force opposing gravity was applied in half of the trials. Pointing errors and different parameters of movement kinematics were analyzed and compared across conditions. It was observed that both the magnitude and the direction of reaching errors were affected by the amount of visual information available during the movement, as well as by the partial gravity compensation provided by the robot.