BACKGROUND Preoperative planning for total shoulder arthroplasty (TSA) may change according to the measured degree of glenoid version. Both 2-dimensional (2D) and 3-dimensional (3D) computed tomographic (CT) scans are used to measure glenoid version, with no consensus on which method is more accurate. However, it is generally accepted that 3D measurements are more reliable, yet most 3D reconstruction software currently in clinical use have never been directly compared to 2D. The purpose of this study is to directly compare 2D and 3D glenoid version measurements and determine the differences between the two. METHODS CT scans were performed preoperatively on 315 shoulders undergoing either anatomic or reverse TSA. 2D measurements of glenoid version were obtained manually using the Friedman method, whereas 3D measurements were obtained using the Equinoxe Planning Application (Exactech Inc.) 3D-reconstruction software. Negative version values indicate retroversion, whereas positive values indicate anteversion. Two observers collected the 2D measurements 2 separate times, and intra- and interobserver measurements were calculated. Groups were compared for variability using intraclass correlation coefficients (ICCs), and for differences in sample means using Student t tests. Additionally, samples were stratified by version value in order to better understand the potential sources of error between measurement techniques. RESULTS For the 2D measurements, intraobserver variability indicated excellent reproducibility for both observer 1 (ICC = 0.928, 95% confidence interval [CI] 0.911-0.942) and observer 2 (ICC = 0.964, 95% CI 0.955-0.971). Interobserver variability measurements also indicated excellent reproducibility (ICC = 0.915, 95% CI 0.778-0.956). The overall 2D version measurement average (-4.9° ± 10.3°) was significantly less retroverted than the 3D measurement average (-8.4° ± 9.1°) (P < .001), with 3D measurements yielding a more retroverted value 73% of the time. When stratified on the basis of version value with outliers excluded, there was no significant difference in the distribution of high-error samples within the data. CONCLUSIONS There was excellent reproducibility between the 2 observers in terms of both intra- and interobserver variability. The 3D measurement techniques were significantly more likely to return a more retroverted measurement, and high-error samples were evenly distributed throughout the data, indicating that there were no discernable trends in the degree of error observed. Shoulder surgeons should be aware that different glenoid version measurement strategies can yield different version measurements, as these can affect preoperative planning and surgeon decision making.