The work aims to establish a rapid, accurate and nondestructive method for testing the plastic bottles of eye drops. Portable differential Raman spectrometer was used to test 33 plastic bottles of eye drops. Then, according to the sample data obtained such as peak position and peak number of the differential Raman spectrum, the composition of the samples was analyzed. At the same time, K-means clustering was carried out to the samples by statistical analysis software "Statistical Products and Services Solutions he most optimal K-means was selected based on Silhouette Coefficient. The samples with a large number in the classification results were analyzed, and the accuracy of the analysis results was verified by Pearson correlation coefficient. The samples were mainly composed of polyethylene and polyethylene terephthalate. The most optimal K-means selected based on Silhouette Coefficient was 2 and the K-means clustering algorithm verified the classification result and composition analysis result of the samples. With polyethylene samples as an example, they could be divided into 7 groups according to different fillers. The calculation results of Pearson correlation coefficient proved the reliability of the above classification and grouping results. This method can accurately and non-destructively detect and classify the plastic bottles of eye drops.