Polycyclic aromatic hydrocarbons (PAHs) are a toxic and harmful organic compound that widely exists in the natural environment. Its main sources are various microorganisms in nature and plant biosynthesis, natural fires in vegetation-rich areas, volcanic eruptions, fossil fuels, incomplete combustion of artificial industrial hydrocarbons and oil leakage during transportation. Polycyclic aromatic hydrocarbons are toxic, with biological carcinogenicity, heritage toxicity and mutagenicity. It harms human respiratory, circulatory and nervous systems in many aspects. Therefore, it is necessary to study the on-site monitoring and analysis methods of polycyclic aromatic hydrocarbons. Chemical analysis methods include chemical titration with pretreatment, liquid chromatography (LC), high-performance liquid chromatography (HPLC), and gas chromatography-mass spectrometry (GC-MS);Spectroscopic analysis studies UV absorption spectra, fluorescence spectra and three-dimensional fluorescence spectra. The three-dimensional fluorescence spectrum obtains the information of excitation and emission wavelength simultaneously, so it contains more optical information, high sensitivity and remarkable spectral characteristics. Therefore, it gains obvious advantages in the field detection of actual water bodies and the rapid study of mixed components of water samples. Common three-dimensional fluorescence spectrum analysis methods include parallel factor analysis (PARAFAC), multidimensional partial least squares (N-PLS), etc. Parallel factor analysis is an effective method to analyze polycyclic aromatic hydrocarbons' overlapping three-dimensional fluorescence spectra. However, sometimes, due to the weak fluorescence signal, the underdetermined three-dimensional fluorescence spectrum analysis cannot get satisfactory results. In order to extract more components from two sample sets, a method based on singular value decomposition(SVD)and PARAFAC is proposed. First, singular value decomposition is used on each observed sample, the appropriate singular value is selected according to the cumulative contribution rate, and a new pseudo sample is constructed to highlight the weak fluorescence signal. They were then inputting two observed samples and their corresponding pseudo samples into PARAFAC to recover the component spectrum. Three groups' three-dimensional fluorescence spectra of polycyclic aromatic hydrocarbons with different fluorescence intensities are analyzed. The pure component spectra of six polycyclic aromatic hydrocarbons were extracted and identified from two mixed samples and the result shows that the similarity between the emission/excitation spectra, and the standard spectra is more than 0.80.