A novel approach to construct theoretical QSAR models is proposed. This tec
hnique, based on the systematic use of quantum similarity measures as theor
etical molecular descriptors, opens the possibility to localize and to iden
tify the position of the bioactive part of drug molecules in situations, wh
ere the nature of the pharmacophore is not known. To test the reliability o
f this new approach, the method has been applied to the study of steroids b
inding to corticosteroid-binding human globulin. The studied molecules invo
lved the set of 31 Cramer's steroids, often used as a benchmark set in QSAR
studies. It has been shown that theoretical QSAR models based on the prese
nt procedure are superior to those derived from alternative existing approa
ches. In addition, a new method to measure the statistical significance of
multiparameter QSAR models is also proposed.