Cast-resin dry-type transformers are widely used in high power applications. Long-term operation increases the extent of damages if no proper safeguards are taken. The major cause of transformer damages is insulation deterioration, which is correlated with partial discharge (PD) occurrence. Recent PD research studies have focused on how to locate the PD occurrence and improve the accuracy of PD detection. The best way to improve the accuracy is to locate the elements affected by the PD. Based on the empirical mode decomposition (EMD) method, the correlated-EMD method (CEMD) was proposed to filter out the uncorrelated intrinsic-mode functions (IMFs), and reveal a PD-affected signal as the filtered result. The spectral CEMD method proposed herein improves the screening process especially for those IMFs with low-frequency vibrations. The results of the spectral CEMD method showed the clearer pattern for PD detection than that of the CEMD method. This algorithm can provide useful information for PD analyses and improve PD identification.