Efficient gene delivery to specified therapeutically relevant cell types within a patient remains a significant challenge in gene therapy. For example, astrocytes outnumber neurons in the human brain, and it is unclear that neurons undergoing insult due to disease or injury should be further burdened by secreting their own neuroprotective factors, making astrocytes an attractive therapeutic target. Efforts to use natural variants of viral vectors have met with limited success. In addition, refractory cell types often possess more than one rate limiting step, such endosomal transport or cell surface binding, and lack of extensive structure/function knowledge of many viruses variants further restricts the use of rational approaches to circumvent these gene delivery barriers. Adeno-associated viral (AAV) vectors have proven to be safe and efficient gene delivery vectors. Previous rational design approaches to design AAV variants has generated vectors with some cell selective gene delivery properties. The AAV capsid structure is a major determinant of its transport and gene delivery properties, and we have therefore developed novel approaches to engineer the capsid proteins to enhance their gene delivery capabilities. Specifically, we have generated a wide range of functionally diverse AAV libraries, through DNA shuffling, error prone PCR, and random peptide sequence insertion into the AAV capsid. Iterative rounds of positive selection of these libraries, involving infection and amplification by addition of adenovirus, coupled with negative selections have allowed us to evolve novel AAV virions with desired phenotypes. For example, we have evolved novel AAV variants with gene delivery efficiencies at least 10-fold higher than the original parent virus on primary human astrocytes and murine astrocytes. Initial characterization of these variants demonstrated elevated levels of cell surface binding relative to the original virus, and viral competition assays suggest the use of an alternate cell surface receptor by the mutants. This work demonstrates that directed evolution provides a powerful approach for engineering viruses with customized gene delivery properties and strong therapeutic potential in vivo.