Detailed information about muscle forces in the human musculoskeletal system are highly demanded for several applications. Unfortunately, the measurement of muscle forces in-vivo is hardly possible. To date, musculoskeletal models are best alternative for the direct measurement of these forces. A major concern in musculoskeletal modeling is, however, model validity. To validate a model we need to compare its predictions to real measured muscle forces which, as mentioned, are difficult to measure. The main objective of this thesis is the validation of a comprehensive musculoskeletal shoulder model, the Delft Shoulder and Elbow Model, at the level of kinematic and dynamic models. Recently, an implantable instrumented shoulder endoprosthesis has been developed that is capable of measuring contact loads in the glenohumeral joint in-vivo. Although direct measurement of muscle forces is still not possible by this implant, it does allow for a general validation at the level of the summed muscle forces in the glenohumeral joint. In the first part of this thesis, the in-vivo measured forces in the instrumented shoulder implant are used, as a ‘golden standard’, to quantitatively validate the model. A general conclusion of this part is that the model is moderately accurate for estimation of glenohumeral joint reaction forces. The second part of this thesis explores the reasons for the differences between model predictions and experimental data that were observed and described in the first part and focuses on model adjustments to find a closer match between model and experiment. A selection of potential causes (e.g. muscle co-contraction, articular friction) is identified and investigated. An important conclusion of this part would be that antagonist muscle co-contraction should be considered in the modeling procedure. It is also concluded that friction moments in the shoulder endoprosthesis are considerable and should be included in the biomechanical analysis of artificial shoulder joints.