A new approach for simultaneous dual-voxel J-difference spectral editing is described, which uses spatially selective spectral-editing pulses and Hadamard encoding. A theoretical framework for spatial Hadamard editing and reconstruction for parallel acquisition (SHERPA) was developed, applying gradient pulses during the frequency-selective editing pulses. Spectral simulations were performed for either one (gamma-aminobutyric acid, GABA) or two molecules (glutathione and lactate) simultaneously detected in two voxels. The method was tested in a two-compartment GABA phantom, and finally applied to the left and right hemispheres of 10 normal control subjects, scanned at 3 T. SHERPA was successfully implemented at 3 T and gave results in close agreement with conventional MEGA-PRESS scans in both the phantom and in vivo experiments. Simulations for GABA editing for (3 cm)3 voxels in the left and right hemispheres suggest that both editing efficiency losses and contamination between voxels are about 2%. Compared with conventional single-voxel single-metabolite J-difference editing, two- or fourfold acceleration is possible without significant loss of SNR using the SHERPA method. Unlike some other dual-voxel methods, the method can be used with single-channel receiver coils, and there is no SNR loss due to unfavorable receive-coil geometry factors.