During the practice of combinatorial chemistry, it has been realized that molecular diversity is not the only essential feature in a synthetically feasible library. In addition, it is of utmost importance to enrich potential libraries with those molecules which could be converted to viable drug candidates. Given the enormous number of potentially synthesizable compounds, there is a need to design a subset of true drug-like compounds. In addition, a paradigm shift in drug discovery has resulted in the integration of pharmacokinetic and drug development activities into early stages of lead discovery. In particular, in silico filters are being developed and used to help identify and screen out compounds that are unlikely to become drugs. This paper highlights recent computational approaches towards the design of drug-like compound libraries, in particular, the prediction of drug-likeness in a more general sense as well as intestinal absorption through passive transport, the permeation of the blood-brain barrier and recent developments towards identification of potentially metabolically unstable molecules. Current computational tools for library design allow the incorporation of medicinal chemistry knowledge into library planning by a variety of methods, ranging from the use of privileged building blocks and simple counting of structural properties (e.g. number of hydrogen bonding partners) to relatively complex regression or neural network-based models to explain oral bioavailability and other pharmacokinetic properties by structural features. These tools are being incorporated more frequently into drug design according to the rule-of-five which refers to simple descriptors correlated to oral drug absorption. Combining experimental knowledge with effective computational filtering and prediction of various aspects of drug-likeness thus facilitates the rapid and cost-effective elimination of poor candidates prior to synthesis and helps focus attention on interesting molecules.