To solve the effect of redundant information caused by high-dimensional Raman-spectrum data on the rapid and accurate identification of the aging state of transformer oil-paper insulation, a Raman-spectrum feature-extraction method for oil-paper insulation based on local linear embedding is proposed. An accelerated thermal-aging experiment was performed to obtain 100 groups of oil-paper insulation samples at different aging stages. The samples were classified into 10 categories based on the polymerization degree of the insulation paper, and Raman spectroscopy was performed on the samples. The conventional principal component analysis (PCA) and locally linear embedding (LLE) feature-extraction methods were used to extract features from the Raman spectrum. Adaboost was introduced to build a discrimination model, and the aging status of the two feature-extraction results was discriminated. Next, the subsequent discrimination accuracy of the two feature-extraction methods was compared. The results show that the discrimination accuracies of the Raman-spectrum samples after feature extraction using LLE and PCA are 98.8% and 90.2%, respectively, which proves that LLE feature extraction offers a greater discrimination accuracy and facilitates subsequent discrimination. The identified spectral information reflects the data simplification and accurate discrimination of oil-paper insulation Raman-spectrum samples via LLE feature extraction combined with the Adaboost discrimination model, which has practical engineering significance for the aging discrimination of transformer oil-paper insulation.