The implantation of genetically engineered nonneuronal cells can provide an effective method for achieving localized delivery of discrete molecules to the CNS or for providing substrates for regrowth of neural structures. Most primary nonneuronal cells have the advantage of being easily obtainable from the prospective host for ex vivo retrovirus-mediated genetic manipulation (most will be mitotic in culture) and reimplantation as an autologous graft (circumventing the problem of immune rejection). As primary cells, they are unlikely to be tumorigenic. The most vexing problem for such systems remains the apparent loss of transgene expression from viral promoters after prolonged periods of engraftment. Much effort is currently being directed at optimizing sustained transgene expression by varying the promoters, by varying the cell types to be engineered, or by regulating expression by enhancing promoter function or substrate availability. While nonneuronal cells are excellent vehicles for achieving passive delivery of substances to the CNS, they lack the ability to incorporate into the host cytoarchitecture in a functional manner (e.g., make synaptic contacts). For this reason, not only may certain essential circuits not be re-formed, but the regulated release of certain substances through feedback loops may be missing. While apparently unimportant for some substances (e.g., ACh), for others (e.g., NGF), their unregulated, inappropriate, excessive, or ectopic release may actually be inimical to the host. Furthermore, the loss of foreign gene expression (the bane of gene therapy) may leave engineered nonneural cells incapacitated, whereas donor tissue originating from brain may intrinsically produce various CNS factors allowing correction to proceed despite inactivation of the introduced gene. In fact, CNS-derived tissue may provide as-yet-unrecognized endogenous neuralspecific substances which are equally as beneficial to the host as the gene in question. Thus, future developments in gene delivery to the brain for some conditions may emphasize using neurons or neural progenitors for ex vivo genetic manipulation (Fisher, 1997) and refining techniques for the direct injection of therapeutic genes into neurons in vivo (see Snyder and Fisher, 1996). For a wide variety of conditions, however, using nonneuronal cellular vehicles or even nonbiologic synthetic vehicles may be efficient, effective, and safe strategies for the passive delivery of therapeutic molecules to discrete regions of the CNS. In fact, this approach may come closer than any other to immediate human applications.