Zhou, J., Ba, J., Castagna, J.P. and Carcione, J.M., 2019. Application of trace-based spectral principal component analysis method for seismic thin-layer thickness delineation. Journal of Seismic Exploration, 28: 551-576. Spectral decomposition of a 3-dimensional reflection seismic volume generates large volumes of spectral data in the form of time-frequency analysis at every seismic signal location. Conventional spectral principal component analysis (PCA) compresses the multi-dimensional spectral data exclusively on amplitude maps at interpreted seismic horizon. This overlooks the time-variant nature of spectral amplitudes. Hence, it is difficult to estimate thin-layer thickness variations directly from the conventional horizon-based spectral PCA (HSPCA) results. A trace-based spectral principal component analysis (TSPCA) method is proposed for seismic thin-layer thickness delineation. Compared to HSPCA, TSPCA calculates spectral principal components (PCs) within a time window over the targeted seismic event on a trace-by-trace basis. Trace-based spectral PCs are assumed independent, i.e., as amplitude responses from reflection events with different frequency characteristics. A rotation of PC coefficients following the Varimax criterion is proposed to automatically interpret the three most significant spectral PCs as related to (1) reflection amplitude determined by rock impedances, (2) tuning of a pure even-reflection pair, or (3) tuning from a pure odd-reflection pair. The latter two types of tuning-related amplitude are both governed by thin-layer thickness and have different frequency responses. Results on synthetic wedge models of pure odd- and even-reflection pair thin layers show that the trace-based spectral PCs show a distinct relationship to thin-layer thickness. Comparing spectral PC images calculated on a geologically complex 3D model after HSPCA and TSPCA methods, we conclude that TSPCA has superior capability for precise thickness delineation, especially for subtle thickness variations in the model.