A variety of computational tools that are used to assist drug design are reviewed. Particular emphasis is given to the limitations and merits of different methodologies. Recently, a number of general methods have been proposed for clustering compounds in classes of drug-like and non-drug-like molecules. The usefulness of this classification for drug design is discussed. The estimation of (relative) binding affinities is from a theoretical point of view the most challenging part of ligand design. We review three methods for the estimation of binding energies. Firstly, quantitative structure-activity relationships (QSAR) are presented. These have gained significantly from recent developments of experimental techniques for combinatorial synthesis and high-throughput screening as well as the use of powerful computational procedures like genetic algorithms and neural networks for the derivation of models. Secondly, empirical energy functions are shown to lead to more general models than standard QSAR, since they are fitted to a variety of complexes. They have been used recently with considerable success. Thirdly, we briefly outline free energy calculations based on molecular dynamics simulations, the method with the most sound theoretical foundation. Recent developments are reestablishing the interest in this approach. In the last part of this review structure-based ligand design programs are described. These are closely related to docking, with the difference that in design, unlike in most docking procedures, ligands are built on a fragment-by-fragment basis. Finally, a short description of our approach to computational combinatorial ligand design is given.