Singular value decomposition (SVD) has a shortcoming in feature extraction of mineral oil′ s 3D fluorescence spectrum, as it easily discards small eigen values that may be important for identification. A new method that combines wavelet transform (WT) and SVD in feature extraction is presented. Wavelet approximation components of mineral oil′s 3D fluorescence data and detail components in different directions are obtained, and their singular value feature is extracted. The fuzzy clustering method (FCM) is used to classify or discriminate mineral oils, and a further test is carried out with random noise introduced. The result shows that WT-SVD feature vector is superior to SVD in mineral oil classification or oil identification, with higher accuracy and robustness than SVD in anti-jamming performance.