<p>Thermal tomographic images could be extensively utilized for their noteworthy usefulness in various applications, specifically in security, civil inspection work, medical, and industrial domains. Accurate and effective classification of thermal tomographic images poses a significant challenge in assessing temperature distribution, like whether they are uniform or not, due to the complex image content and the scarcity of annotated datasets. This paper proposes some multivariate statistics-based techniques to excerpt features from thermal tomographic images and thereby their classification towards evaluation of heating patterns in a closed contour inside a hot air-flow chamber. These images are reconstructed using COMSOL Multiphysics software based on Finite Element Method for various boundary sensor measurements as a result of different heating and fluid flow patterns maintained inside the said chamber. Multivariate statistical techniques used in this article involve the Eigenface Euclidean distance, Principal Component Analysis (PCA) based similarity, Histogram intersection similarity and combined Skewness-Kurtosis distance and Chi-square statistic metrics for the said classification purpose. This work achieves satisfactory classification accuracies of 100%, 77.78%, 100%, 77.78% and 88.88% using the Eigenface Euclidean distance, PCA-based similarity factor, histogram intersection similarity index, combined skewness-kurtosis distance metric and Chi-square statistic, respectively while employed on our small thermal tomographic image dataset. A comparative analysis of these methods to other methods available in literature has also been presented using two standard thermal image datasets. This kind of indirect estimation of heating quality may be employed for hot fluid-flow processes.</p>