Purpose: To evaluate spectral acquisition processes important for obtaining reliable and reproducible γ-aminobutyric acid (GABA) signals from volunteers in brain regions that are frequently used for neuroimaging studies [anterior cingulate cortex (ACC), superior temporal gyrus, and caudate] at ultra-high field. Methods: Ten healthy volunteers were studied using a single-voxel Point-RESolved Spectrosocpy (PRESS) sequence with band selective inversion with gradient dephasing pulses (BASING). The editing pulse was designed to be symmetrically placed at 2.0 and 1.4 ppm in the two cycles to reduce the co-editing of macro-molecules (MM). Spectral data were obtained with phase encoding matrix 8 × 8 × 1 and two editing cycles or 1 × 1 × 1 and 64 editing/64 non-editing. The total acquisition time was approximately 4.5 min for each acquisition. An automated MRS prescription method was utilized for the placement of the GABA scan location in 5/10 subjects. Three regions of interest were predefined in the MNI152 space and then registered and transformed to subject space. These volunteers also had repeat scans to examine between-session reproducibility. Results: The placement of editing pulses symmetrically at 1.7 ppm reduced the effect of MM contributions and provided more accurate GABA estimation. Chemical shift misregistration errors caused by classic PRESS localization sequence are more significant at ultra-high field strength. Therefore, a large over-excitation factor was needed to reduce this error. Furthermore, the inefficiency of saturation bands and unspoiled coherence could also interfere with the quality of the data. Reliable recovery of metabolite signals resulted from the implementation of 8 × 8 × 1 phase encoding that successfully removed artifacts and errors, without compromising the total acquisition time. Between successive scans on the same subject, dice overlap ratios of the excited spectral volume between the two scans were in the range of 92-95%. Within subject variability of metabolites between two repeat scans was smaller in the ACC and left superior temporal gyrus when compared to that in the right caudate, with averaged coefficients of variation being 3.6, 6.0, and 16.9%, respectively. Conclusion: This study demonstrated the feasibility of obtaining reliable and reproducible GABA measurements at ultra-high field.