As drug discovery and development has grown ever riskier and more expensive, interest has increased in using computational tools to identify good candidates more quickly and to avoid investing resources in synthesizing and testing compounds that are not likely to succeed. The most powerful of these tools seek to exploit the connection between the three-dimensional (3D) structure of a molecule and its various biological activities. Two fundamentally different ways of addressing this challenge have arisen over the years: ligand-based methods that seek to identify and exploit similarities between the structures of ligands that are known to bind to a given target; and target-based (docking) methods that seek to identify and exploit complementarities to the binding site itself. The structure-activity relationships involved reflect the interplay of many thermodynamic factors, which makes putting them on any kind of quantitative footing a challenge. The progress being made in coming to grips with that challenge is assessed here through a survey of recently published prospective studies, a review of the underlying problems remaining, and consideration of some ways in which different methods are being combined to improve overall performance.