We describe a differential wax-wane focus servo technique for use in optical data storage. A combination of scalar diffraction modeling and experiment is used to quantify performance. Our results indicate that the differential technique is superior to the single-detector wax-wane technique with respect to gain, linearity, and lock-on range. We present modeling results that show the effects of aberrations and detector misalignment. The differential system was found to be robust. It can also reject many common pattern noise effects such as tracking cross talk, which was reduced from 0.7 µm in a single channel to < 0.1 µm in the optimized differential channel.