With the continuous development of petroleum resources in the ocean, more and more petroleum is leaking into the marine environment. It not only threatens the marine ecological environment but also seriously affects people’s health.Therefore, the rapid and effective detection of petroleum pollutants in the marine environment is of great significance for the protection of the marine ecological environment and human health.Petroleum products contain a large number of polycyclic aromatic hydrocarbons, which have strong fluorescence characteristics.Therefore, fluorescence spectroscopy technology has become one of the important means to detect petroleum pollutants. In this paper, three-dimensional fluorescence spectroscopy combined with parallel factor analysis algorithm and pattern recognition method is used to characterize and classify petroleum pollutants. Firstly, the micelle solution prepared by seawater and sodium dodecyl sulfate (SDS) was used as a solvent to prepare different concentrations of diesel,jet fuel, gasolineand lube solutions, and 80 experimental samples were finally obtained. Then, three-dimensional fluorescence spectra of experimental samples were collected by FLS920 fluorescence spectrometer, and the effect of scattering was removed by using the Delaunay triangle interpolation method. Secondly, the paralleled factor analysis (PARAFAC) algorithm is used to decompose the three-dimensional fluorescence spectrum data after scattering, and the component number is estimated by using the nuclear consistency diagnosis method and residual analysis method. Finally, in order to establish a robust classification model, 80 experimental samples were divided into 60 training set samples, and 20 test set samples by Kennard-Stone algorithm.The K-nearest neighbor (KNN) algorithm, principal component discriminant analysis (PCA-LDA) algorithm and partial least squares discriminant analysis (PLS-DA) algorithm are used to establish the classification model respectively, and sensitivity, specificity and accuracy are used to evaluate the classification effect.The results show that the recognition accuracy of the three classification models is 85%, 90% and 94% respectively. The PLS-DA classification model has the highest recognition accuracy and the best classification effect.Therefore, based on extracting the fluorescence spectrum data of petroleum pollutants by using parallel factor analysis algorithm and combining with the pattern recognition method, the classification of different kinds of oil products can be well studied.In this paper, three-dimensional fluorescence spectroscopy combined with parallel factor analysis algorithm and pattern recognition method is used to detect petroleum pollutants quickly and effectively, which provides a new research idea and an important reference for the rapid detection of petroleum pollutants.