A number of preclinical studies have shown the adeno-associated virus (AAV) to be an efficient vehicle for gene therapy. Clinical studies successfully demonstrated its potential for in vivo gene transfer. The complexity of host–vector interactions when progressing from small to large animal models, and eventually to humans, has impeded translation of AAV technology to the clinic. One approach to address this complexity has been to explore the biological characteristics of variations in AAV capsid structure. Initial strategies characterized the naturally occurring capsid variants from mammalian species. The structural and functional knowledge gathered on these natural AAV variants as vectors has led to the first series of second-generation vectors that aim at specifically improving certain properties by rational design of the capsid. A third exciting approach uses directed evolution to isolate vectors that are able to overcome selective pressures applied in the laboratory and thereby steer the capsid to evolve toward improved functionality.