Three-dimensional fluorescence spectroscopy plays an important role in studying the fluorescence information of polycyclic aromatic hydrocarbons (PAHs). PAHs are carcinogenic and refractory. They are mostly produced by exhaust emissions and waste incineration, which endanger human health and the environment. Therefore, people are constantly exploring the detection methods of PAHs. ANA and NAP in PAHs were selected as detection substances and FLS920 fluorescence spectrometer was used in the experiment. In order to avoid the influence of Rayleigh scattering produced by the fluorescence spectrometer itself, the initial emission wavelength was set at 40 nm, and the excitation wavelength was lagged behind, and the scanning excitation wavelength (lambda ex) was set at 200～370 nm, and the emission wavelength (lambda em) was set at 240～390 nm. Then we could gain the fluorescence data of PAHs obtained by fluorescence scanning, and we could analyze ANA and NAP qualitatively and quantitatively in mixed solution by the three-dimensional fluorescence spectroscopy and PARAFAC. The ANA and NAP used in the experiment were purchased from the Aladdin reagent official website, and we prepared a stock solution with a concentration of 10 mg·L-1, and we should dilute the stock solution, and we canget 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5 mg·L-1 of secondary stock solution, which obtain a concentration of ANA and NAP, Then we maxed the solution of ANA and NAP. Before spectral analysis, the spectra of ANA and NAP needed to be pretreated, and we should eliminate the effect of Raman scattering by blank subtraction method, and adopt the way of ensemble empirical mode decomposition (EEMD) to eliminate interference noise. In this experiment, there are two peaks in ANA, located at λex=298 nm, λem=324/338 nm, and the peaks of NAP at λex=280 nm and λem=322 nm. The PARAAFAC algorithm selected in this paper was very sensitive to the choice of component number, therefore, using the method of nuclear consistency diagnosis to estimate the number of components, and the nuclear consistency values of the estimated values 2 and 3 were all over 60%, then decomposed the mixed samples by PARAFAC of 2 and 3 factors respectively. After decomposition, the data of excitation emission spectra and concentration of each component were normalized, and we can draw the spectrogram, and compare with the real excitation emission spectrogram and concentration map of each component. At the same time, the predicted concentration of mixed samples obtained by PARAFAC was used to determine the accuracy of quantitative analysis by calculating the recovery rate (R) and root mean square error (RMSEP). When choosing two factors, the fitness of ANA and NAP was 95.7% and 96.7%, the average recovery was 101.8% and 98.9%, the root mean square error was 0.018 7 and 0.031 6, and choosing three factors, the fitness of ANA and NAP was 95.3% and 95.8%, the average recovery was 97% and 102.5%, the root mean square error was 0.033 and 0.116. Because of the three indicators, the effect of qualitative and quantitative analysis with two factors was better than that with three factors. The experimental results showed that the qualitative and quantitative analysis of mixed samples based on three-dimensional fluorescence spectrometry and PARAFAC algorithm can effectively determine the type of mixed samples, and its can successfully predict the concentration of mixed samples.