Classical single-neuron recording methods led to 'neuron-centric' concepts of neural coding, whereas more recent multi-neuron population recordings have inspired 'population-centric' concepts of distributed processing in neural systems. Because most neocortical neurons code information coarsely, sensory or motor processing tends to be widely distributed across neuronal populations. Dynamic fluctuations in neural population functions thus involve subtle changes in the overall pattern of neural activity. Mathematical analysis of neural population codes allows extraction of 'motor signals' from neuronal population recordings in the motor cortices, which can then be used in real-time to directly control movement of a robot arm. This technique holds promise for the development of neurally controlled prosthetic devices and provides insights into how information is distributed across several brain regions.