Stress affects a significant proportion of the world’s population, making it a crucial topic to study. This work explores the use of facial thermographic imaging to automatically identify thermal response patterns associated with stress. Ten university students underwent a short Trier Social Stress Test (TSST), during which infrared thermal images were captured. Subsequently, the temperature was extracted from key regions of interest (ROIs) in the facial area, including the forehead, tip of the nose, cheeks and the maxilla. A hierarchical clustering model method was then applied, using the histogram from these regions as feature sets to group participants according to their physiological thermal responses. The Ward linkage method was applied to compute the pairwise distances and generate a dendrogram that visually and quantitatively grouped participants according to similarities in their thermal response profiles. The results show a trend towards a mind-body relationship, however, the correlation between clustering and self-reported stress levels was not statistically significant (p = 0.166). Nevertheless, the cluster structure revealed physiologically significant clusters, suggesting inter-individual differences in thermal responses to stress. These results demonstrate the potential of unsupervised learning methods as a means of generating labels based on physiological data rather than subjective perception. This reduces labelling bias and provides an additional approach to stress detection in affective computing and psychophysiological research.