Action is the means by which we and animals survive. It consists of a complex combination of movements which are either innately endowed or acquired by learning. In this article, I propose a hypothesis on the relationship between the organization of action and the organization of the brain. Innate and learned actions are controlled by different levels of neural networks: innate actions are controlled by reflex mechanisms and pattern generators in the spinal cord and brainstem, while learned actions are controlled by the cerebral cortex, basal ganglia, and cerebellum. However, these mechanisms are by no means independent. Phylogenetically, animals have acquired progressively more complex actions by gaining neural connections between different neural mechanisms. This is accomplished by the connection from newly evolving brain structures, particularly the cerebral cortex, to reflex or pattern generator mechanisms, as typically observed in the neural mechanism for saccadic eye movement. The cerebral cortex is a general purpose device which can be used in different ways depending on biological demands; in other words, it is used for learned actions. In consequence, a given movement (e.g., saccade) can be produced by different neural circuits, all converging onto the movement generation mechanism (e.g., s.c.) in an excitatory manner. However, such converging inputs that promote actions are likely to produce a chaotic explosion of neural signals. There must be some way to prevent the explosion and select signals that are most appropriate for the current behavioral context. The basal ganglia system evolved to accomplish this goal. It exerts a powerful inhibition on its targets in the brainstem (e.g., s.c.) and the thalamo-cortical system, thereby closing the gate for the action-promoting excitatory inputs; it also removes the sustained inhibition using another inhibition originating in the striatum (input structure of the basal ganglia), thereby opening the gate so that an appropriate action is executed. There are at least two additional functions of the basal ganglia. First, the selection mechanism of the basal ganglia is used also for the selection of simulated actions (e.g., thoughts) which are largely controlled by the association cortices. Second, it is used for learning of behavioral procedures: various kinds of signals from the cerebral cortex converge onto neurons in the basal ganglia to generate temporary association of neural signals, whose behavioral significance is evaluated by signals from the limbic system via dopaminergic neurons. The procedural memories thus created (perhaps in the cerebral cortex, particularly premotor cortices) are then used to guide learning of individual movements in which the cerebellum plays a crucial role. Thus, the implementation of learned actions is carried out by two distinct neural systems, each forming a loop circuit: 1) cerebral cortex and basal ganglia; 2) cerebral cortex and cerebellum. Although these neural systems are independent structurally, they work in parallel and cooperatively to acquire and execute learned procedures (actions).