A cerebrospinal fluid (CSF) to brain exchange has been postulated for lipid-soluble and small molecular substances and has led to nearly 100 attempts per year to produce central effects via intrathecal injections. With few exceptions, however, modern neurological practice has avoided this approach because of its demonstrated ineffectiveness and dangers. The practicability of an intrathecal CSF to brain exchange was tested by cisternal infusions of mock CSF at different infusion pressures that might counteract central nervous system intoxications of systemic origin. Those efforts failed in different test situations with each of three barbiturates. Steady state doses at a selected level of barbiturate anesthesia were the same, whether induced by cisternal infusion or intravenously, and this was true for barbiturates of widely different lipid solubility. The cerebral response to pentylenetetrazol was delayed well beyond its rate of response when introduced intravenously. These results suggested that the bulk clearance rate and venous resorption of CSF were sufficient to prevent significant diffusion of the barbiturate or even mock CSF into the brain following intrathecal injection. Because central effects that follow venous resorption may be confused with direct central effects, many previous clinical reports are questioned. Apparent exceptions to the ineffectiveness of intrathecal therapy, such as spinal anesthesia, were discussed in terms of their special local effects. The relative effectiveness of intrathecal agents should be evaluated by comparing maintenance doses for a given central effect, when produced by both intrathecal and i.v. routes. Previous reports on rates of intrathecal infusion, intracranial pressure relationships, and the relative safety of such infusions were confirmed and extended.