The multiple linear regression (MLR) analysis and back propagation neural networks (NN) were performed to examine the quantitative structure-polarization relationships (QSPR) for the formation of antibody-BTEX-EDF complex. Five descriptors out of 18 ones were selected for both MLR and NN, respectively, and the selected descriptors in MLR were the same as those in NN. These descriptors were the number of atoms, which can form hydrogen bonds (HA), connolly surface area (Area), the highest occupied molecular orbital energy (HOMO), partial charge of C3 carbon atom (C3), and HOMO pi coefficient of C2 carbon atom (P2). The fact that the descriptors in MLR are identical to those in NN suggests that these descriptors have good linear relationships and play a significant role in the formation of antibody-tracer complex.