Local field potentials (LFPs) in the brain are an important source of information for basic research and clinical (i.e., neuroprosthetic) applications. The energy contained in certain bands of LFPs in the 10-100 Hz range has been shown to correlate with specific arm movement parameters in nonhuman primates. In the near future, implantable devices will need to transmit neural information from hundreds of microelectrodes, and transcutaneous data transfer will become a significant bottleneck. Here we present a low-power, fully-integrated circuit that performs on-site data reduction by isolating LFPs and measuring their signal energy. The resulting analog VLSI circuit consumes 586 microm x 79 microm of silicon area and dissipates only 5 nanowatts of power. We show that the chip performs similarly to state-of-the-art signal processing algorithms.