OBJECTIVE: Until recently, magnetocardiography (MCG) studies were performed using SQUID systems, consisting of a planar array of sensors with uniform spacing. The introduction of optically-pumped magnetometers (OPMs) now enables the deployment of large, conformal arrays, in which the sensors can be mounted on a wearable vest at nearly any location. The objective of this study was to optimize the sensor array geometry of an OPM system for MCG imaging applications. METHODS: We devised a new optimization criterion for spatial resolution based on sensitivity to localization error. We also implemented a greedy optimization technique to overcome the difficulty of combinatoric optimization over an extremely large number of possible sensor configurations. Simulations were performed to compare the localization accuracy of the optimized arrays to that of conventional arrays with a regular geometry over the front of the torso. The number of sensors and the signal-to-noise ratio were varied. RESULTS: Optimization resulted in non-planar, irregular geometries biased toward the left half of the torso. Arrays optimized for posterior cardiac sources showed the best overall performance. The localization accuracy was shown to be significantly improved by optimization for a given number of sensors and signal-to-noise ratio. CONCLUSION: The results of this study can serve as a guide for designing MCG arrays for a given number of sensors and/or determining the required number of sensors for a given level of performance. SIGNIFICANCE: Sensor array optimization can improve the performance of OPM-based MCG imaging systems for applications, such as non-invasive localization of arrhythmogenic foci.