Structure activity relationships for a series of TIBO (4,5,6,7-tetrahydro-5-methylimidazo[4,5,1-j,k][1,4] benzodiazepin-2(1H)-one) derivatives, which significantly inhibit HIV-1 replication are analyzed by the Electron-Topological Method (ETM) and Artificial Neural Networks (ANNs). Activities of the TIBO series including 91 compounds are given as IC(50). Conformational analysis and quantum-chemical calculations are carried out for each TIBO derivatives, and then molecular fragments being specific for active compounds and non-active compounds are revealed by using ETM. In this study, we used optimized geometry data and electronic characteristics to form Electron-Topological Matrices of Contiguity (ETMCs) for all compounds in the series of TIBO derivatives. Effective charges on atoms are taken as diagonal elements, bond characteristics and optimized distances represent non-diagonal elements. To obtain the algorithmic base for the activity prediction, ANNs were used after the ETM (the so-called combined ETM-ANN method). As the result, 6 pharmacophores and anti-pharmacophores were chosen as the most important ones. The statistical coefficients calculated by the proposed algorithm were q(2)=0.82, for training set and q(2)= 0.72, for external test set respectively Thus, the found results showed that ETM-ANNs approach is a good convenient tool for QSAR studies.