How the elaborate neuronal circuit in the cerebellum operates and is involved in motor learning is a question addressed in earnest in studies on the cerebellum. During the past four decades, experimental studies have revealed circuit and module structures of the cerebellum, established long-term depression (LTD) as a unique and characteristic type of synaptic plasticity in the cerebellum, and analysed signal contents of activates of cerebellar neurons related to motor learning. In the 1990s, these studies were developed to detailed analyses of the signal transduction underlying LTD, and to uncovering the involvement of the cerebellum in cognitive function. On the other hand, theoretical studies yielded epochal Marr-Albus network models of the cerebellum around 1970, and introduced control system principles explaining the essential roles of the cerebellum in motor learning as providing internal models, both forward and inverse. The author maintains the hypothesis that reorganisation of the neuronal circuit by error-driven induction of LTD constitutes the major memory and learning mechanisms of the cerebellum. In this article, we examine the validity of the hypothesis in light of currently available data in recent studies of the cerebellum.