Although peripheral prosthetic systems for the central nervous system, such as retinal and cochlear implants, have existed for some time and unidirectional brain computer interface systems that map physiological states into states of external devices have already been realized (Horch et al,2004), the bidirectional coupling of artificial systems with the central nervous system that can replace specific functionality has thus far not been demonstrated. In order to realize such a system three fundamental problems must be overcome. First, the function of the circuit to be replaced must be understood and captured in a real-time form. Second, the inputs and outputs to and from the circuit that is replaced must be identified and their signals correctly analyzed and synthesized. Third, steps 1 and 2 must be physically realized in a small, efficient and low-power form that can support implantation. Here we demonstrate our success in addressing the first two challenges by describing the offline real-time coupling of simultaneously recorded signals from the input structures of an anesthetized rat cerebellum, the pons and the inferior olive, to a computational model of the cerebellum (Verschure et al, 2001) that is implemented in real-time hardware (FPGA). We demonstrate that this prototype of a component of the neuroprosthetic system can partially substitute the function of a cerebellar microcircuit in eye blink conditioning and acquire correctly timed conditioned responses to auditory conditioned stimuli and can maintain the acquired memory trace for hours. Hence, our results provide the first proof of concept of the feasibility of implantable self-contained neuroprosthetic systems that can partially replace the input-output function of neuronal circuits.