Intracranial microelectrode arrays (MEAs) are ubiquitous tools used for recording brain activity. During recordings, a fraction of channels in an array can be inactive for various reasons. The inclusion of data from inactive channels compromises the accuracy of any technique that pools data from multiple channels. Modern MEAs can contain hundreds to thousands of channels, and the inactive channels may change from experiment to experiment, rendering manual identification impractical. Therefore, we have developed a technique to automatically and rapidly identify active channels based on the presence of brain waves, quantified in terms of power spectral density from 4–100 Hz. We applied this technique to MEA recordings in a non-human primate as it performed 120 trials of a directional joystick task, and found that decoding accuracy (as measured by mean correlation coefficient) increased from 0.38 to 0.63 after inactive channels were identified and removed. Our method constitutes a novel technique that can be used to improve the fidelity of MEA output, for applications such as brain-machine interfaces and neurophysiology.