For phenolic acids in cosmetics, some are added as active ingredients, such as caffeic acid with skin-repairing effects, gallic acid capable of anti-inflammatory and anti-allergic, etc.; some are added as preservatives, such as p-hydroxybenzoic acid, sorbic acid, etc.; some are prohibited substances illegally added by bad businesses, such as hydroquinone, resorcinol and so on. In order to monitor the quality of cosmetics, the detection of phenolic substances in cosmetics is particularly important. Many researchers have done related work for this purpose. The method of separation and analysis based on chromatography has achieved certain success, but the disadvantages such as being time-consuming and costly and complicated operation are also obvious. Three-dimensional fluorescence spectroscopy has a high sensitivity, but fluorescence interference and spectral overlap have a large impact on detection, and complex cosmetic samples often fail to achieve the desired results. In order to realize simultaneous qualitative and quantitative detection of phenolic acids in cosmetics, the paper combines three-dimensional fluorescence spectroscopy with four-dimensional calibration of chemometrics (also called third-order correction) to overcome unknown interference and the effect of collinearity with the data while ensuring high sensitivity. First, select in the linear range of caffeic acid (CA), p-hydroxybenzoic acid (p-HA), hydroquinone (HQ). At the appropriate concentration, calibration samples, verification samples and cosmetic samples were prepared at the four pH values of 7.00, 7.30, 7.50, 7.80, respectively, so that the excitation-emission-pH-sample (EX-EM-pH-Sample) four-dimensional data were obtained. Secondly, in order to verify the effect of pH on the fluorescence intensity, 320 nm was chosen as the excitation wavelength to obtain the emission wavelength of caffeic acid at four pH values. It was found that the fluorescence intensity of caffeic acid increased with the increase of pH value, indicating the introduction of pH. The value was reasonable as the fourth dimension. Finally, the appropriate component number was selected to decompose and predict the four-dimensional data matrix by alternating penalty quadrilinear decomposition (APQLD). The decomposed spectrum was compared with the actual spectrum, and the predicted concentration was compared with the actual concentration. The experimental results showed that the decomposition spectrum can be consistent with the actual spectrum whether it is a validation sample or a cosmetic sample. The average recovery (AR) of the validation sample was 100.4% to 103.5%, and the predicted root mean square error (RMSEP) was less than 0.06. The average recovery (AR) of cosmetics sample was 100.0%~102.2%, and the predicted root mean square error (RMSEP) was less than 0.08. Compared with chromatographic studies, the recovery rate increased by about 4%, and the operation was simple, time-saving and labor-saving, and has high sensitivity. Compared with the second-order correction method, multiple components in the complex cosmetic system can be realized under unknown interference. Replacing “physical and chemical separation” with “mathematical separation” is fast, efficient, economical and environmentally friendly; and third-order correction can overcome certain data collinearity problems and improve sensitivity to some extent.