Once sufficient efficacy is obtained, the success or failure of a potential drug depends on its absorption, distribution, metabolism, excretion and toxicity (ADMET) characteristics. Since a tremendous amount of effort, for example high throughput screening and synthesis, has gone into finding bioactive molecules, ADMET issues are often the rate limiting factors in drug discovery programs. In response a significant amount of effort has gone into developing in silico models for a large variety of ADMET issues. The progress in several of these areas, blood-brain barrier penetration (Klon), protein binding (Hall), hERG inhibition (Diller), and hepatotoxicity (Cheng), are reviewed in this special issue. In addition, the limitations and best uses of in silico ADMET models are discussed in two additional reviews (Franklin, DeLisle). One particular challenge, mentioned in each of the reviews, facing anyone wishing to develop an in silico ADMET model is obtaining a large amount of consistent and appropriate data over a wide range of chemical structures. First, nearly all biological end points are the combination of multiple phenomena. Blood brain barrier penetration can occur by passive diffusion or active transport (Klon). In addition, it can be limited by being actively exported. Further, simple penetration into the central nervous system might not be sufficient to obtain efficacy. Ultimately, the free fraction in the brain will determine whether sufficient compound is available to achieve the desired efficacy. On the surface, binding to serum proteins or blockage of the hERG potassium channel are seemingly single phenomenological endpoints, but even these endpoints are more than simply binding to their respective proteins. Plasma protein binding involves multiple proteins and multiple binding sites within each protein (Hall). In addition, the kinetics of binding can be quite important whereas equilibrium binding is usually just considered. To bind to the hERG channel pore, the likely site of binding for most hERG blockers, a molecule must first pass through the cell membrane and then enter the channel through the activation gate (Diller), thus, cell permeability is an important piece to understanding hERG blockers. Of all the endpoints, hepatotoxicity is clearly the most complex and challenging. The hepatotoxic potential of a compound likely arises through many different mechanisms, is highly dependent on compound concentration, and is highly dependent on metabolites and intermediates thereof. Of course, metabolism itself can occur through many pathways making predicting the hepatoxic potential of a compound that much more difficult. Even if a large data set were available from a single source, one still must ask the question whether any one in vitro assay format for a particular end point is clearly superior to another. Assays invariably change over time because they are not perfect predictors of in vivo or clinical results and are tuned as more information becomes available. Thus a model fit too heavily to data from a particular assay might win the battle - better predictions for that assay - but lose the war - poorer predictions for in vivo or clinical effects. Clearly the ultimate goal of in silico models is to provide guidance on the in vivo effects (or more specifically, in human effects) of assessed compounds, so one must choose in vitro endpoints carefully keeping in mind their relationship to the in vivo goal.