The measurement of electrophysiology is of critical importance to our understanding of brain function. However, current non-invasive measurements—electroencephalography (EEG) and magnetoencephalography (MEG)—have limited sensitivity, particularly compared to invasive recordings. Optically-Pumped Magnetometers (OPMs) are a new type of magnetic field sensor which ostensibly promise MEG systems with higher sensitivity; however, the noise floor of current OPMs remains high compared to cryogenic instrumentation and this limits the achievable signal-to-noise ratio of OPM-MEG recordings. Here, we investigate how sensor array design affects sensitivity, and whether judicious sensor placement could compensate for the higher noise floor. Through theoretical analyses, simulations, and experiments, we use a beamformer framework to show that increasing the total signal measured by an OPM array—either by increasing the number of sensors and channels, or by optimising the placement of those sensors—affords a linearly proportional increase in signal-to-noise ratio (SNR) following beamformer reconstruction. Our experimental measurements confirm this finding, showing that by changing sensor locations in a 90-channel array, we could increase the SNR of visual gamma oscillations from 4.8 to 10.5. Using a 180-channel optimised OPM-array, we capture broadband gamma oscillations induced by a naturalistic visual paradigm, with an SNR of 3; a value that compares favourably to similar measures made using conventional MEG. Our findings show how an OPM-MEG array can be optimised to measure brain electrophysiology with the highest possible sensitivity. This is important for the design of future OPM-based instrumentation.