This work describes the use of a multilevel Plackett-Burman design (PB) for the creation of a calibration set for partial least square regression (PLS). The PB calibration set was compared to a collinear analogue by testing these two PLS models for the analysis of six polycyclic aromatic hydrocarbons (PAHs). These compounds were analyzed in micellar media by synchronous fluorescence after determination of the experimental conditions (choice of surfactant, analytical conditions such as deltalambda, step, and scan range). The external validation shows that the collinear set is inappropriate to quantify PAH in real samples, but the PB calibration set affords optimal results.