Abstarct: Electromayogram signals known as electrical pulses are sent to muscular fibers by human neural system. These signals are useful in muscle behavior assessment and have some clinical applications.Today, there is a great tendency to transmit and store long-term EMG recordings. Moreover, along with the need for suitable methods for compression EMG signals, we should put in mind the compression should be done in such a way that useful clinical information is preserved after compression and thus, medical diagnosis is still fully reachable. In this thesis, we have proposed two EMG signal compression approach: first one baxsed on Empirical-Mode-Decomposition baxsed signal approximation and Discrete-Cosine-Transform baxsed signal smoothing and second one baxsed on pre/de-emphasis technique. In first proposed approach the role of EMD method is approximation and signal relative smoothing and also providing the ability to control the quality of signal approximation. DCT is also used to smooth the EMG signal which inherently has a relatively high frequency behavior. In second proposed, we have used the pre/de-emphasis technique in the Fourier domain to produce a smooth signal from the input EMG signal. In each tow approach the smoothness signal is compressed by wavelet transform and SPIHT coding after bidimensionality. The proposed methods are evaluated by two sets of criteria measuring the compression efficiency (including the PRD and CF measures) and capability of preserving the clinical information (including four spectral parameters). baxsed on the results, for compressive values of 75%, 80%, 85%, 90 %, in first approach the PRD values are equal to 0.82, 1.55, 2.49, 4.8 for Isometric Contraction signals and 1.24, 2.67, 4.12, 6.25 for dynamic signals and in second approach PRD values are equal to 0.94, 1.72, 2.53, 4.96 for Isometric Contraction signals and 1.36, 2.83, 4.25, 6.36 for dynamic signals that show the superiority of the proposed methods compared to existing approaches.