The evolution of drug resistance in HIV has been a major obstacle in combating the AIDS epidemic, and development of the next generation of antiviral drugs will depend on improvements in the methodology addressing resistance. This work examines HIV evolution from a structural perspective, focusing on the development of methods to anticipate drug resistance and improve drug discovery efforts. To understand the evolution of HIV in the presence of inhibitors requires knowledge of viral replication capacity as well as drug resistance. Replication capacity can be predicted with a phylogenetic approach, which estimates impairment in HIV protease activity. Pairing these estimates with a structural model based on molecular docking allows the detection of most major clinically observed protease mutations. Combining structural modeling and analysis of existing protease mutations generates predictions of drug resistance mutations for an experimental protease inhibitor. Mutagenesis experiments validate these predictions, while also revealing epistatic interactions and cross-resistance with existing inhibitors. A fitness model based on predicted replication capacity and drug resistance is able to rank in vitro HIV mutant infectivity with significant accuracy. This fitness model is incorporated into a simulation of viral evolution, which correlates with clinically observed mutation prevalence. Simulations also affirm the high level of cross-resistance among protease inhibitors, highlighting the importance of alternative drug targets. Current drug discovery projects often use computer-based models of protein-ligand interaction for docking and virtual screening. A novel analysis of binding energy results from large-scale virtual screening identifies representative wild-type and mutant protease structures, focusing future efforts. Complementary efforts in the study of APS reductase reveal correlations between the distribution docking results and the underlying energy surface. Cluster analysis is shown to be an empirical measure of docking entropy which can improve the accuracy of binding energy predictions. Applying these insights in a virtual screen for new inhibitors of HIV protease, a library of 1,585 compounds is narrowed to 36 candidates. Five of these compounds prove to be inhibitors. Modeling indicates that two of them bind outside the protease active site, suggesting potential leads for a new class of protease inhibitor