In this paper, a novel feature extraction algorithm for power quality (PQ) disturbance signal classification is proposed based on extracting spectral features from Discrete Cosine Transform (DCT) domain. The spectral domain feature extraction offers the ability to detect and localize transient events and thereby classify different power quality disturbance signals or events. For optimal feature set selection, a novel technique of selecting significant DCT coefficients is proposed, which in addition to offering feature dimensionality reduction, results in high within-class-compactness and between-class-separation. For the classification purpose, an Euclidean distance-based classifiers has been employed upon the proposed feature space. Seven types of PQ disturbance signals have been considered and extensive simulations have been carried out, which show that the extracted features provide a very high classification accuracy at a low computational burden, even with a very simple distance-based classifier.