Modeling studies were performed on HCV NS5B polymerase in an effort to design new inhibitors. The binding models of five different scaffold inhibitors were investigated and compared by using molecular dynamics simulations, free energy calculation and decomposition. Our results show Tyr448 plays the most critical role in the binding of most inhibitors. In addition, favorable contributions of residues Pro197, Arg200, Cys366, Met414 and Tyr448 in a deep hydrophobic pocket prove to be important for the selectivity of inhibitors. Furthermore, an optimized docking protocol was presented based on cross-docking the five inhibitors in the palm binding site of this enzyme using the Autodock program. This protocol was used later to virtually screen NCI and Maybridge diversity set libraries. The binding site was profiled via the statistics and analysis of the hydrogen bond networks formed between the receptor and the top-ranked diversity set compounds. Based on our detailed binding site analysis two useful rules were proposed to guide the selection of promising hits.