Analysis of the partial discharge signal, a technique to diagnose the stator winding insulation is a key function for the diagnosis of high voltage rotating machines and requires high precision. To satisfy this requirement, various denoising techniques such as filtering and differential methods were proposed. However, these techniques can not eliminate a adjacent induction noise that decreases reliability of the diagnosis. A simple novel denoising algorithm, therefore, is proposed for removing the adjacent induction noise in this paper. The algorithm shows good performance in the real partial discharge signals measured by 13kV class capacitive couplers installed at hydro-generator in Dae-cheong Dam.