Glenohumeral osteoarthritis is a degenerative shoulder joint disease, which causes the breakdown of articular cartilage and bone. People affected by this disease suffer from severe pain and eventually reduced mobility and poor life quality. This disease is unfortunately prevalent in the elderly population. In most advanced cases, a total joint replacement or total shoulder arthroplasty (TSA) is needed. Anatomical Total Shoulder Arthroplasty (aTSA) is usually performed in cases of intact rotator cuff muscles. It involves the replacement of the damaged cartilage and exposed bones by prostheses, thereby re-establishing the relative motion of the joint and dramatically reducing pain. As a result, basic human functions such as lifting a coffee pot or combing one's hair are restored. Although the aTSA is usually an excellent procedure, its failure rate is relatively high mostly because of the aseptic glenoid implant loosening, often associated to an unstable loading on the glenoid implant. This instability may be reinforced by the presence of glenohumeral subluxation, defined as the relative position of the humeral head with respect to the glenoid fossa. A new design of prosthesis, called overcorrected prosthesis (OC), was proposed as a means to reduce the postoperative subluxation; however, it was never tested experimentally nor clinically. The objective of this thesis was to evaluate the potential of posterior OC implants on the reduction of posterior subluxation of patients planned for aTSA. Thus, a patient-specific numerical model of the shoulder bone was developed in order to compare standard versus OC implants. This numerical finite element (FE) model was based on patients of the Lausanne University Hospital (CHUV) who were planned for aTSA, and required three main steps: First, based on the patients' pre-surgery computed tomography (CT) data, the glenoid bone FE models of each patient was developed. Virtual implant placement was performed for each FE model, replicating preoperative planning and matching postoperative CT. Second, patient-specific loading was computed thanks to a patient-specific musculoskeletal model (MSM) during activities of daily living (ADL). Third, this FE model required a material law for the bone in order to predict the bone reaction to the implant design. First, an experimental set-up involving Digital Volume Correlation (DVC) was designed: cadaveric implanted glenoids were loaded in a micro-CT compatible device and scanned before and during loading. DVC was used to extract displacement and compressive strain at the peri-implant area with enough precision to be used in numerical validation studies. Second, five specimen-specific FE models were created which replicated the experiment. While the displacement was very well replicated (coefficient of determination R2 = 1.0 and slope =1.0), the strain measurements were poorly replicated (R2 = 0.28-0.37 and slope = 0.51-0.70). The workflow and the most suitable material law found in this part of the thesis were used to create patient-specific FE models that compared standard and OC implants. Results indicate that the OC design has the potential of reducing postoperative subluxation without over constraining the underlying bone. It would be interesting to extend this study to a larger population in order to confirm the potential advantage of the OC implants.