Cognitive Radio has become an effective solution for the scarcity and the inefficient use of the spectrum. The performance of the cognitive radio (CR) system depends however on how well it can maximize the usage of the free bands without causing interference to primary users. The ability to detect the primary user in CR systems relies on Spectrum Sensing (SS). In this paper we propose a method to improve the performance of Energy Detection SS using Wavelet Denoising. Cooperative and non-cooperative sensing scenarios are considered. Results show that sensing accuracy can be highly improved in the cooperative scenario when the leaves of the tree of wavelet packet transform are used as a basis for denoising. Different mother wavelets are also tested. Simulation results are analyzed and are compared to traditional results.