Efficient virus and non-virus vector systems for gene transfer to tumor cells have been developed and tested in cell culture and in animal experiments. With some of the earliest and most comprehensively evaluated vectors, such as retroviruses, advanced clinical trials were performed in tumor patients. Malignant primary brain tumors (gliomas) have been chosen for the first clinical studies on novel gene therapy approaches because these tumors are non-metastatic and develop on the largely postmitotic background of normal glial and neuronal tissue. However, the human cancer gene therapy studies performed so far were not as successful as preclinical animal experiments. Furthermore, the clinical studies did not address major limiting factors for in vivo gene therapy, such as insufficient gene transfer rates to the tumor with the used local delivery modalities, and the resulting inability of a particular transgene-prodrug system to confer permanently eradicating cytotoxicity to the whole neoplasm. Critical evaluation of gene transfer and therapy studies has led to the conclusion that, even using identical vectors, the anatomical route of vector administration can dramatically affect both the efficiency of tumor transduction and its spatial distribution, as well as the extent of intratumoral and intracerebral transgene expression. This review concentrates on different physical methods for vector delivery to malignant primary brain tumors in experimental or clinical settings: stereotactic or direct intratumoral injection or convection-enhanced bulk-flow interstitial delivery; intrathecal and intraventricular injection; and intravascular infusion with or without modification of the bloodtumor- barrier. The advantages and drawbacks of the different modes and delivery routes of in vivo vector application, and the possibilities for tumor targeting by modifications of the native tropism of virus vectors or by using tissue-specific or inducible transgene expression are summarized.