It has been shown in earlier works that the vertices of Platonic solids are good measurement choices for tests of Einstein-Podolsky-Rosen (EPR)-steering using isotropically entangled pairs of qubits. Such measurements are regularly spaced, and measurement diversity is a good feature for making EPR-steering inequalities easier to violate in the presence of experimental imperfections. However, such measurements are provably suboptimal. Here, we develop a method for devising optimal strategies for tests of EPR-steering, in the sense of being most robust to mixture and inefficiency (while still closing the detection loophole, of course), for a given number n of measurement settings. We allow for arbitrary measurement directions, and arbitrary weightings of the outcomes in the EPR-steering inequality. This is a difficult optimization problem for large n, so we also consider more practical ways of constructing near-optimal EPR-steering inequalities in this limit.