Drug-discovery projects frequently employ structure-based information through protein modeling and ligand docking, and there is a plethora of reports relating successful use of them in virtual screening. Hit/lead optimization, which represents the next step and the longest for the medicinal chemist, is very rarely considered. This is not surprising because lead optimization is a much more complex task. Here, a homology model of the histamine H(3) receptor was built and tested for its ability to discriminate ligands above a defined threshold of affinity. In addition, drug safety is also evaluated during lead optimization, and "antitargets" are studied. So, we have used the same benchmarking procedure with the HERG channel and CYP2D6 enzyme, for which a minimal affinity is strongly desired. For targets and antitargets, we report here an accuracy as high as at least 70%, for ligands being classified above or below the chosen threshold. Such a good result is beyond what could have been predicted, especially, since our test conditions were particularly stringent. First, we measured the accuracy by means of AUC of ROC plots, i. e. considering both false positive and false negatives. Second, we used as datasets extensive chemical libraries (nearly a thousand ligands for H(3)). All molecules considered were true H(3) receptor ligands with moderate to high affinity (from microM to nM range). Third, the database is issued from concrete SAR (Bioprojet H(3) BF2.649 library) and is not simply constituted by few active ligands buried in a chemical catalogue.