Principal component analysis was employed to identify the sources of polycyclic aromatic hydrocarbons soil contaminants, and then an artificial neural network was applied to investigate the contribution of the polycyclic aromatic hydrocarbons to soil biotoxicity. An artificial neural network model with a 4-9-1 architecture provided the most accurate results in terms of correlation coefficient (r, 0.7592), root-mean-square error (0.7396), and bias (-0.0879). The contamination sources included traffic emissions (13.2%), petroleum-related products (37.4%), coal combustion (36.0%), and combustion of other fossil fuels (13.4%).