Sensory guidance of behavior often involves standard visuo-motor mapping of body movements onto objects and spatial locations. For example, looking at and reaching to grasp a glass of wine requires the mapping of the eyes and hand to the location of the glass in space, as well as the formation of a hand configuration appropriate to the shape of the glass. But our brain is far more than just a standard sensorimotor mapping machine. Through evolution, the brain of advanced mammals, in particular human and non-human primates, has acquired a formidable capacity to construct non-standard, arbitrary mapping using associations between external events and behavioral responses that bear no direct relationship. For example, we have all learned to stop at a red traffic light and to go at a green one, or to wait for a specific tone before dialing a phone number and to hang up when hearing a busy signal. These arbitrary associations are acquired through experience, thereby providing primates with a rich and flexible sensorimotor repertoire. Understanding how they are learned, and how they are recalled and used when the context requires them, has been one of the challenging issues for cognitive neuroscience. Valuable insights have been gained over the last two decades through the convergence of multiple complementary approaches. Human neuropsychology and experimental lesions in monkeys have identified a network of brain structures important for conditional sensorimotor associations, whereas imaging studies in healthy human subjects and electrophysiological recordings in awake monkeys have sought to identify the different functional processes underlying the overall function. The present review focuses on the contribution of a network linking the prefrontal cortex, basal ganglia, and dorsal premotor cortex, with special emphasis on results from recording experiments in monkeys. We will first review data pointing to a specific contribution of each component of the network to the performance of well-learned arbitrary visuo-motor associations, as well as data suggesting how novel associations are formed. Then we will propose a model positing that each component of the fronto-striatal network makes a specific contribution to the formation and/or execution of sensorimotor associations. In this model, the basal ganglia are thought to play a key role in linking the sensory, motor, and reward information necessary for arbitrary mapping. (C) 2004 Elsevier Ltd. All rights reserved.