Discrete wavelet transform has become a widely used feature extraction tool in pattern recognition and pattern classification applications. However, using all wavelet coefficients as features is not desirable in most applications -- the enormity of data and irrelevant wavelet coefficients may adversely affect the performance. Therefore, this paper presents a novel feature extraction method based on discrete wavelet transform. In this method, Shannon's entropy measure is used for identifying competent wavelet coefficients. The features are formed by calculating the energy of coefficients clustered around the competent clusters. The method is applied to the lung sound classification problem. The experimental results show that the new method performs better than a well-known feature extraction method that is known to give the best results for lung sound classification problem.