Fluorescence spectrum properties of benzo [k] fluoranthene (BkF), benzo[a]pyrene (BaP), and their mixture are analyzed. The experimental results show that there are two characteristic fluorescence peaks existing in BkF and six characteristic fluorescence peaks in BaP. The optimal emission wavelength of BkF and BaP is 405 nm . The fluorescence characteristics of the mixture vary widely when there are different concentration ratios. But the optimal emission wavelength of the mixture with different concentration ratios does not change, remaining at 405 nm. When the excitation wavelength is 250~400 nm and the emission wavelength is 350~500 nm, serious overlapping occurs in the fluorescence spectra of BkF and BaP. In order to detect their concentrations in the mixture precisely, back propagation (BP) neural network optimized by artificial bee colony (ABC) algorithm is applied. The result indicates that the ABC-BP neural network is better than the genetic algorithm (GA)-BP method. The ABCBP neural network can accurately detect the concentration of BkF and BaP in their mixture when the concentration is in the range of 1.000~10.000 ng/L. The average recovery of BkF and BaP in ten mixture groups is 99.19% and 99.26%, respectively.