Grip force control relies on accurate internal models of the dynamics of our motor system and the external objects we manipulate. Internal models are not fixed entities, but rather are trained and updated by sensory experience. Sensory feedback signals relevant object properties and mechanical events, e.g. at the skin-object interface, to modify motor commands and update internal representations automatically. Here we prove that intact sensory feedback is essential for predictive grip force regulation. The efficiency and precision of grip force adjustments to load fluctuations arising from vertical and horizontal point-to-point arm movements with a hand-held object were analysed in a chronically deafferented subject (G.L.) and three healthy control subjects. Point-to-point movements started and ended with the object being held stationary. G.L. and healthy controls produced similar accelerations of the grasped object and consequently similar load magnitudes during vertical and horizontal movements. Compared with healthy controls, G.L. employed inefficiently high grip forces when holding and moving the object, indicating inaccurate force scaling to object weight and inertial loads. For healthy controls, the grip force profile was precisely timed to the movement-induced load fluctuations during vertical and horizontal movements. However, G.L.'s grip force profile was not processed to match differential loading requirements of movement direction. We conclude that predictive grip force control requires at least intermittent sensory feedback to signal the effectiveness of descending motor commands and to update internal models.