Virtual screening is an important resource in the drug discovery community, of which protein–ligand docking is a significant part. Much software has been developed for this purpose, largely by biochemists and those in related disciplines, who pursue ever more accurate representations of molecular interactions. The resulting tools, however, are very processor-intensive. This paper describes some initial results from a project to review computational chemistry techniques for docking from a non-chemistry standpoint. An abstract blueprint for protein–ligand docking using empirical scoring functions is suggested, and this is used to discuss potential improvements. By introducing computer science tactics such as lazy function evaluation, dramatic increases to throughput can and have been realized using a real-world docking program. Naturally, they can be extended to any system that approximately corresponds to the architecture outlined.