Partial discharge (PD) monitoring can reveal various types of insulation defects and provide an effective means for condition assessment of high voltage (HV) equipment. One of the challenging tasks of PD monitoring is to effectively extract PD signals from acquired signals, which are susceptible to extensive noise. The extracted PD signals can then be further processed for identifying the insulation defects that generate the discharges, i.e. PD pattern classification. This paper proposes an eigenvector algorithm (EVA)-based blind equalization (BE) for PD signal de-noising. An original noise-corrupted PD signal is processed by EVA to generate a series of equalized signals. By using kurtosis as a selection criterion, an optimal equalized signal that recovers PD impulses can be obtained. Pulse sequence of the optimal equalized signal can be used for PD pattern classification. To verify the effectiveness of the proposed de-noising method, extensive laboratory experiments were conducted. The results show that the proposed method is capable of extracting PD signals from original signals, which are overwhelmed by severe background noise. Moreover, the extracted signals preserve pulse sequence patterns, which are distinctive for different insulation defects and consistent without affected by the types of PD sensors and data acquisition systems.