The problem of low signal-to-noise ratio for gamma-aminobutyric acid (GABA) in vivo is exacerbated by inefficient detection schemes and non-optimal experimental parameters. To analyze the mechanisms for GABA signal loss of a MEGA-PRESS J-difference sequence at 4 T, numerical simulations were performed ranging from ideal to realistic experimental implementation, including volume selection and experimental radio frequency (RF) pulse shapes with a macromolecular minimization scheme. The simulations were found to be in good agreement with phantom and in vivo data from human brain. The overall GABA signal intensity for the simulations with realistic conditions for the MEGA-PRESS difference spectrum was calculated to be almost half of the signal simulated under ideal conditions (~43% signal loss). In contrast, creatine was reduced significantly less then GABA (~19% signal loss). The 'four-compartment' distribution due to J-coupling in the PRESS-based localization was one of the most significant sources of GABA signal loss, in addition to imperfect RF profiles for volume selection and editing. An alternative strategy that reduces signal loss due to the four-compartment distribution is suggested. In summary, a detailed analysis of J-difference editing is provided with estimates of the relative amounts of GABA signal losses due to various mechanisms. The numerical simulations presented in this study should facilitate both implementation of the more efficient acquisition and quantification process of J-coupled systems.