We have developed a new ligand based refinement strategy and program, DynaDock, for homology models. The algorithm performs an iterative refinement of a given homology model by alternately docking known inhibitors into the receptor (with FlexX) and refining the docked complexes by a new combined Monte-Carlo and molecular dynamics approach. Next to the refinement of the binding pocket side chains, which is the focus of most other approaches, our algorithm allows the adjustment of the overall backbone conformation in the vicinity of the binding pocket. DynaDock was developed and extensively evaluated (docking of known inhibitors and virtual screening) during a drug design project, which produced several highly potent and selective lead compounds for the inhibition of aldosterone synthase (CYP11B2). We have extended the program for general use and evaluated it through the successful refinement of homology models of various known (X-Ray) cytochrome P450 structures (improvements in the C_a-RMSD up to 3.2 Å).