Dihydrofolate reductase (DHFR) is an important therapeutic target for treatment of cancer and microbial disease. Its species specificity has resulted in the sequencing of a number of vertebrate and bacterial DHFRs, and the three-dimensional structure of isozymes from Escherichia coli, Lactobacillus casei, and chicken liver has been elucidated, in the presence of the coenzyme NADPH and of a number of inhibitors. This information has enabled scientists to try to design improved and more selective inhibitors, based on the known coordinates of the enzyme features. Simple use of computer graphics or wire models has resulted in the design of inhibitors with 50 times the activity of trimethoprim, an antibacterial DHFR inhibitor, by making use of an unused ionic binding site. However, in a number of instances this approach was completely unsuccessful because hydrophobic sites of interaction were preferred. More sophisticated techniques involve energy minimization of the small molecule-macromolecule interactions to optimize the geometry. In this paper I describe the use of a molecular mechanics program, AMBER, for predicting the geometry and relative energetics of binding. Very encouraging results have been obtained for a closely related series of compounds. Where differing entropic and solvent effects are involved, predictions may be poor. The use of super computers and molecular dynamics methods should increase this capability in the near future.