Restoring the premorbid proximal humeral anatomy during shoulder arthroplasty is critical yet can be difficult because of the deformity of the arthritic head. The purpose of this study was to measure the variation between surgeons and between types of prosthetics in reproducing the anatomic center of rotation (COR) of the humeral head after anatomic shoulder arthroplasty. The anteroposterior radiographs of 125 stemmed and 43 resurfacing shoulder arthroplasties, performed by 5 experienced surgeons, were analyzed. All patients had primary replacement for treatment of end-stage glenohumeral arthritis. A best-fit circle to preserved nonarticular humeral landmarks was used to define the difference between the anatomic COR and the prosthetic COR. A difference in COR of >3.0 mm was considered clinically significant and analyzed for the cause of this deviation. The average deviation of the postoperative COR from the anatomic COR was 2.5 ± 1.6 mm for stemmed cases and 3.8 ± 2.1 mm for resurfacings. Thirty-nine stemmed cases (31.2%) and 28 resurfacings (65.1%) were beyond 3.0 mm of deviation and regarded as outliers. The majority of the stemmed outliers and all resurfacing outliers were overstuffed. An improper humeral head size selection and inadequate reaming were the main reasons for the deviation in stemmed and resurfacing outliers, respectively. A large percentage of shoulder replacements demonstrated significant deviations from an anatomic reconstruction. Resurfacing arthroplasty exhibited significantly greater deviations compared with stemmed arthroplasty (P < .001), indicating that surgeons have more difficulty in restoring the anatomy with resurfacings. Further studies are needed to assess the clinical impact of these deviations.