With the vigorous development of the astronomical spectral big data acquired, such as LAMOST, assessments of the automated data reduction and analysis are necessary. The above work uses the Euclidean distance analysis to determine the similarity between LAMOST spectra and the template. The accuracy of star classification depends on the high-quality template spectra. Classification results from LAMOST 1D pipeline depend on the 183 templates, of which the dependencies should be inspected. In this paper, we calculate both Euclidean and Mahalanobis distances for each pair of templates, using these methods to get the template mean and maximum of A, F, G, K, M stars’. By completing the correlation analysis, we find that the distances averagely show similarity except for several templates. The Mahalanobis distances can even detect the difference between adjacent pairs. They can further identify that the slight differences between the similar templates have better discriminating effects. We conclude from our experiment that most of the LAMOST spectra are correctly classified, while some outstanding templates should be checked as the basis of the optimization for improving the accuracy and reliability of LAMOST templates.