The quantitative structure tribo-ability relationships (QSTR) between 20 triazine derivatives and their respective extreme-pressure properties as lubricant additives were analyzed by the back propagation neural network (BPNN) method. The BPNN-QSTR model had satisfactory stability and predictive ability (R2=0.9965, R2(LOO)=0.9195, q2=0.8274). The anti-wear model also yielded good predictions (R2=0.9757, R2(LOO)=0.6261, q2=0.8022). Two- and three-dimensional structural descriptors were used to analyze molecular structures that affected extreme-pressure and anti-wear properties. The three-dimensional molecular dimensions and the bonding modes of the skeleton atoms in the molecules were important factors. In addition, the effects of N, P, O, and other hetero-atoms on the tribological properties were reflected in their corresponding group types and electronic structures.