The Free-Wilson approach was applied to two groups of catamphiphilic multidrug resistance (MDR) modifiers using classical multiple linear regression (MLR) and genetic algorithms (GA) for feature selection. In the first group (17 thioxanthenes) the side chain length between the ring system and tertiary nitrogen, the type of the tertiary nitrogen substituent and the stereoisomery were found to be significant for anti-MDR activity both by MLR and GA (r(2) = 0.803, predictive power Q(2) = 0.652). In the second data set (17 phenothiazines and related drugs) the ring system type, the stereoisomery, the side chain type, and the ring substituent kind in position two contributed significantly (r(2) = 0.938 and Q(2) = 0.908). The QSAR studies showed a thioxanthene ring with a -CF3 substituent in position two, a piperazine moiety with a 4-bond distance from the ring system and trans-isomery to be optimal for MDR reversal. Based on these results molecular modeling of trans-(T) and cis-flupentixol (C) was performed assuming that the 2 to 3-fold difference in MDR reversing activity of T compared to C might be related to different preferable conformations in the membrane lipid environment. Among all conformations generated by the SYBYL systematic search routine those comprising local energy minima were selected and optimized with semiempirical quantum chemistry methods. The optimized conformations were compared with H-1-NMR analysis results on drug conformations in lipid environment, some of them corresponded excellently. The electrostatic and lipophilic fields of T and C were compared to identify molecular properties related to the activity difference. The results demonstrated that T and C could have a different (mirrorlike) orientation entering the lipid bilayer by the ring system suggesting much better fitting of T compared to C to the lipid ''MDR-reversal receptor''.