Recent advances in the field of chemometrics have provided us with an opportunity to determine if it is possible to extend numerical multivariate pattern recognition techniques beyond simple fuel characterization, to the identification of important compositional features that are related to fuel quality. This can include unique combinations of normally benign constituents that exert an impact on fuel stability. Fuels are ideal candidates for chemometric analysis, because we are often concerned with minute features within a complex compositional matrix. Two potential benefits of this approach are the development of diagnostic and predictive models that can relate fuel composition to quality. We have begun our investigation with studies of gas chromatography-mass spectrometry (GC-MS) data from fuels that have undergone various levels of thermally induced autoxidation. An analysis of variance (ANOVA) feature selection technique has been applied to locate features in the GC-MS data that change from sample to sample, thus allowing for a quick evaluation of how fuel composition is altered during stress. In this manner, evaporative losses, rather than fuel degradation, have been observed to dominate the chemical variations that are produced in naval distillate fuels (NATO F-76) during oven stress at 60 degreesC. Thermal stress in a closed low-pressure reactor (LPR) eliminates evaporative losses, and the chemical changes have been readily observed and modeled. A progressive change in composition during both oven and LPR stress is revealed from multi-way principal component analysis. Decomposition of windowed regions of the GC-MS data via parallel factor analysis provides a means of extracting the mass spectra of individual fuel constituents that change during stress. This illustrates the potential diagnostic capability of multi-way chemometric analysis of GC-MS data.