Estimation of muscle forces in over-actuated musculoskeletal models involves optimal distributions of net joint moments among muscles by a standard load-sharing scheme (SLS). Given that co-contractions of antagonistic muscles are counterproductive in the net joints moments, SLS might underestimate the co-contractions. Muscle co-contractions play crucial roles in stability of the glenohumeral (GH) joint. The aim of this study was to improve estimations of muscle co-contractions by incorporating electromyography (EMG) data into an upper limb musculoskeletal model. To this end, the model SLS was modified to develop an EMG-assisted load-sharing scheme (EALS). EMG of fifteen muscles were measured during arm flexion and abduction on a healthy subject and fed into the model. EALS was compared to SLS in terms of muscle forces, GH joint reaction force, and a stability ratio defined to quantify the GH joint stability. The results confirmed that EALS estimated higher muscle co-contractions compared to the SLS (e.g., above 50 N higher forces for both triceps long and biceps long during arm flexion).