Résumé : Although considerable progress has been made in developing models of cerebellar function in sensorimotor control, the exact nature of the basic operations performed by the cerebellum remain elusive. Several major theories have emerged these last decades. According to the hypothesis of Marr and Albus, the climbing fiber input carries an error signal weakening the strength of a subset of parallel fibers/Purkinje neurons synapses in the cerebellar cortex. Cerebellar circuits would gain the control of movement through trial and error. The hypothesis of internal models emulating movements is currently highly cited. There is a general agreement that (1) the central nervous system has to cope with an intrinsic time delay of sensory feedback related to motor activities and (2) estimations of future motor states are essential to perform fast and accurate movements. According to this second theory, cerebellar dysmetria, one of the cardinal cerebellar deficits, would result from a distorted predictive control. A third popular theory relates to the inverse models that would be stored in the cerebellum. Acquisition of a motor act would require forward models, and the acquisition process itself would generate an inverse model to allow an unconscious coordinated movement. Recently, an international panel of experts from various disciplines discussed the prevailing opinions in a consensus statement and tried to extract their clinical relevance in terms of pathogenesis of the clinical symptoms. Although a consensus is still not reached, the prevailing opinions provide a sound framework to conduct novel studies and try to discover the secrets of cerebellar circuits. © 2013 Springer Science+Business Media New York.