Abstract: The main objective of this study aims to precisely classify the cracks and cavities in surface and sub-surface by using features-based giant-magneto-resistive pulsed eddy current (GMR-PEC) sensor. A new defect feature named as the power spectral density analysis of the direct differential PEC response is carried out based-on the amplitude spectrum. Principal component analysis is designed to reduce the dimensional index with the ability of supplying the lower dimensional feature. The PCA combined linear discriminatary analysis (PCA-LDA) and the Bayesian classifier (PCA-Bayes) are both applied for defect classification. Consequently, the experimental results demonstrate that the cracks and cavities in surface and sub-surface can be classified satisfactorily by the proposed methods using the new feature, which have the potential for gauging automatic in-situ inspection for PEC.