Friction modeling is essential for joint dynamic identification and control. Joint friction is composed of a viscous and a dry friction force. According to Coulomb law, dry friction depends linearly on the load in the transmission. However, in robotics field, a constant dry friction is frequently used to simplify modeling, identification and control. That is not accurate enough for joints with large payload or inertial and gravity variations and actuated with transmissions as speed reducer, screw-nut or worm gear. A new joint friction model taking dynamic and external forces into account is proposed in this paper. A new identification process is proposed, merging all the joint data collected while the mechanism is tracking exciting trajectories and with different payloads, to get a global LS estimation in one step. An experimental validation is carried out with a prismatic joint composed of a Star high precision ball screw drive positioning unit.