Neuro-musculoskeletal (NMS) modeling based on electromyographic (EMG) signals allows for a better understanding of the principles of human movement [1]. Through appropriate modeling software (CEINMS) it is possible to estimate the muscular forces and joint moments calibrated on the single individual. The possibility of having highly subject-specific models allows an improvement in the treatment of a great variety of neurodegenerative diseases and can be fundamental for the general understanding of the neuro-mechanics of human locomotion. Fifteen walking trials of four able body subjects (mean age = 60 ± 2.16, mean BMI = 26.66 ± 4.13) were acquired with a motion capture system (BTS, 60 Hz), synchronized with a force plate (Bertec, 960 Hz) and a 15 channels surface electromyography (sEMG) system (FreePocketEMG, 1000 Hz). The CAST protocol was adopted as marker set, while the following muscles were acquired: Gluteus Maximus, Gluteus Medius, Tensor Fasciae Latae, Sartorius, Adductor Longus, Rectus Femoris, Biceps Femoris, Semitendinosus, Vastus Lateralis, Vastus Medialis, Tibialis Anterior, Peroneus Longus, Gastrocnemius Lateralis, Gastrocnemius Medialis and Soleus. The data were processed in MatLab for the extraction of four muscle synergies by applying the ‘Non-negative Matrix Factorization’ method [2]. Then, 'Statistical Parametric Mapping' [3] and 'Cosine Similarity' were used as metrics to assign each synergy to the EMG signals of just one muscle for each different muscle groups (Ankle Plantar-flexors, Ankle Dorsi-flexors, Knee Extensors and Knee Flexors). Afterwards data were processed in OpenSim; scaling, inverse kinematics, inverse dynamics, muscle analysis were run. Finally, CEINMS was executed to calculate muscle activations and torques first by informing the model with a complete setup of experimental EMGs data then with a reduced setup (the foru EMGs that best matched with muscle synergies). Activation signals of the muscle-tendon units and the angular moments of the knee and ankle obtain from models developed by CEINMS were compared for different setup with the experimental EMG signals. Comparison was done for both different muscle setups for each subject. In Fig. 1a are shown the values of cosine similarities of the muscles able to better approximate each primitive for one subject. In Fig. 1 b-e are displayed the pattern of primitives and the chosen experimental EMG signals for the same subject during one walking trial. The implemented methodology allowed the analysis of the muscular forces developed during walking with an acquisition protocol that includes a reduced number of EMG signals, and a sensitivity analysis for the adoption of NMS models based on a limited number of inputs was conducted. Results seem promising but a validation on a wider sample of subjects including pathological subjects is needed.