Magnetic localization systems based on passive permanent magnets (PM) are of great interest due to their ability to provide non-contact sensing and without any power requirement for the PM. Medical procedures such as ventriculostomy can benefit greatly from real-time feedback of the inserted catheter tip. While the effects of the number of sensors on the localization accuracy in such systems has been reported, the spatial design of the sensor layout has been largely overlooked. Here in this paper, a framework for determining an optimal sensor assembly for enhanced localization performance is presented and investigated through numerical simulations and direct experiments. Two approaches are presented: one based on structured grid configuration and the other derived using Genetic Algorithms. Simulation results verified by experiments strongly suggest that the layout of the sensors not only has an effect on the localization accuracy, but also has an effect far more pronounced than improvements brought by increasing the number of sensors.