Electromyography (EMG) signal is interfered with different kinds of noise and wavelet denoising algorithm is a powerful method to reduce noises in EMG signal. Hard and soft shrinkage, traditional wavelet transformation, are applied to wavelet coefficients with threshold value. From the limitation of hard and soft shrinkage, this study proposes nine improved wavelet shrinkage methods that achieve a compromise between two standards. EMG signal from six hand motions with additive noise at different signal-to-noise ratios were applied to evaluate the efficiency of the methods in denoising viewpoint. In addition, features of estimated denoising signal are sent to classification task to measure the performance in myoelectric control. The experimental results show that adaptive wavelet shrinkage method (ADP) provides the better performance than traditional methods and other modified methods in both of denoising and pattern recognition viewpoints. Accuracy of recognition of EMG signal transformed by ADP is improved about 6.5-78.5% depending on the level of noise. ADP is an efficient method for producing useful EMG signal without noise and improving application of myoelectric control.