Our recent investigations are focused to develop dynamic models for skeletal muscle force and finger angles for prosthetic hand control using surface electromyographic sEMG as input. Since sEMG is temporal and spatially distributed and is influenced by various factors, muscle fatigue and its related sEMG becomes of importance. This study is an effort to spectrally analyze the sEMG signal during progression of muscle fatigue. The sEMG is captured from the arms of healthy subjects during muscle fatiguing experiments for dynamic and static force levels. Filtered sEMG signal is segmented in five parts with 75% overlap between adjacent segments. The analysis is done using different classical (fast Fourier transform, Welch's averaged modified periodogram), model-based (Yule-Walker, Burg, Covariance and Modified Covariance autoregressive (AR) method), and eigenvector methods (Multiple Signal Classification (MUSIC) and eigenvector spectral estimation method) in frequency domain. Results show that the classical and eigenvector based methods are more sensitive than the model-based methods to fatigue related changes in sEMG signals.