---

We present in this paper a novel instrumentation set-up able to interface in real-time a biological neural network (cell culture grown on MEA) with a sensory-motor apparatus. The system was specifically designed (hardware and software) to compute a real-time data loop and proposes on-line and off-line processing functions to help the exploration of the neural network learning capabilities. After presenting the set-up architecture and data flow, we detail its bio-ware, hardware and software parts. The acquisition/stimulation loop timing is performing enough to leave a sufficient time slot (more than 50 μs / sample) for additional user-defined processing functions. We show that using this set-up, we will be able to explore the generic relation between network activity evolution through time and the fact that consistent feedback is provided to it.