Water is the source of life, which is indispensable to people’s diurnal production and life. In recent years, water contamination has become increasingly severe, which has endangered humanity health. Phenolic compounds are organic pollutants that are widespread and difficult to degrade, which refers to hydroxyl-containing derivatives produced by hydroxyl substitution of hydrogen atoms in benzene rings of aromatic hydrocarbons. They are highly toxic to animal, plant and human life activities. Resorcinol (RES) and hydroquinone (HYD) were selected as research object of experiment, and phenol (PHE) was added to three groups of predicted samplesas interference. The samples and blank solvent were scanned by the laboratory FS920 steady-state fluorescence spectrometer to obtain fluorescence spectrum data. The influence of Raman scattering would be eliminated through the deduction of blank solvents method. The obtained data contain the important information in the original spectrum with the greatest extent is preserved while eliminating interference. The spectrum becomes smoother and the fluorescence intensity raise significantly, so the spectral information is more accurate after correction. Second-order correction methods: parallel factor analysis (PARAFAC) and alternating penalty trilinear decomposition (APTLD) together with three-dimensional fluorescence spectroscopy (EEM) were used to measure RES and HYD in form of qualitative and quantitative analysis fast, directly and accurately in two cases: under interference and without interfering stuff meanwhile excitation-emission spectra overlap severely. Because PARAFAC algorithm is sensitive to the component number (i. e. chemical rank) of the mixture system, when the component number is too large the algotithem will arise: falling into the “swamp”, iterations number increases more time consuming. In this paper, Core Consistency Diagnosis (CORCONDIA) is used to estimate component number precisely, which ensures algorithm calculating faster and more accurate. As qualitative analysis results showed that PARAFAC can accurately distinguish RES and HYD without interference. The peak position of RES and HYD are very close, thus it is difficult to distinguish them via traditional methods and “second-order advantage” of combining three-dimensional fluorescence spectroscopy with second-order calibration is demonstrated. The results of quantitative analysis give out that the accuracy of this method is slightly reduced and RMSEP value increases slightly in the presence of interference, but these two methods can still complete the determination accurately. The recovery rate of PARAFAC is 93.4%±0.5%~97.1%±1.0% and the predicted root mean square error is less than 0.190 mg·L-1. Manwhile, the recovery rate of APTLD is 95.9%±1.6%~97.2%±0.8% and the predicted root mean square error is less than 0.116 mg·L-1. By comparing the performance of the above methods, we know PARAFAC is sensitive to the number of components in the samples and strictly linear to the decomposed spectral data. However APTLD has obvious advantages: being insensitive to the number of components, fast calculation speed, strong anti-noise ability, stable results, all of which highlight its advantages.