The classification of ancient glass is of great significance for understanding cultural exchange. However, a large number of chemical compositions of ancient glass, the small variation of its chemical compositions, and the difficulty of obtaining data make it difficult for most classification methods to achieve great results when classifying glass. The PCA-K-means algorithm has been proven to perform well when classifying small-scale data with high-dimensional. In this paper, we propose a weighted Euclidean distance and a composite distance evaluation function to improve the PCA-K-means algorithm further after fully considering the characteristics of the data. We use the novel PCA-K-means algorithm to classify a batch of ancient Chinese glassware compared to traditional K-means and PCA-K-means. The results show that our improved model is better than the other two models in classifying small-scale glass samples.