The phase-encoding strategy plays a critical role in determining the quality of gradient-and spin-echo (GRASE) images. Phase-encoding methods developed for two-dimensional GRASE imaging strive to achieve a balance between artifacts from T2-dependent signal amplitude modulations and off-resonance-dependent signal phase shifts, although no current method provides smooth and nonperiodic evolutions for both of these signal changes. In three-dimensional GRASE imaging, the use of two phase-encoding directions presents the opportunity for improved phaseencoding strategies. In this report a phase-encoding strategy for three-dimensional GRASE, termed SORT, is described; this strategy separates off-resonance and T2 effects, mapping one along each of the two phase-encoding directions. Thus, off-resonance-induced artifacts can be minimized while eliminating T2-dependent periodic signal modulations and allowing complete flexibility in the selection of echo time. The performance of the SORT phaseencoding method for T2-weighted GRASE imaging was compared with that of existing methods based on calculated point spread functions and simulated images. The predicted performance of SORT phase encoding was verified experimentally using T2-weighted three-dimensional GRASE imaging of the brain. Generally artifact-free images were obtained even in the presence of fat, susceptibility interfaces, and a wide range of T2 values. J. Magn. Reson.