Background. The optimal penetration of antiretroviral agents into the central nervous system may be a balance between providing adequate drug exposure to inhibit human immunodeficiency virus (HIV) replication while avoiding concentrations associated with neuronal toxicities. Methods. Cerebrospinal fluid (CSF) exposure of efavirenz and the metabolites 7-hydroxy (7OH) and 8-hydroxy (8OH) efavirenz were assessed after at least 12 weeks of therapy in HIV-infected subjects randomized to commence antiretroviral regimens containing efavirenz at either 400 mg or 600 mg once daily. Results. Of 28 subjects (14 and 14 on efavirenz 400 mg and 600 mg, respectively), CSF HIV RNA was undetectable in all. Geometric mean CSF efavirenz, 7OH-, and 8OH-efavirenz concentrations (with 90% confidence intervals [CIs]) for the 400-mg and 600-mg dosing groups were 16.5 (13-21) and 19.5 (15-25) ng/mL; 0.6 (.4-.9) and 0.6 (.4-1) ng/mL; and 5.1 (4.0-6.4) and 3.1 (2.1-4.4) ng/mL, respectively. Efavirenz concentration in CSF was >0.51 ng/mL (proposed CSF 50% maximal inhibitory concentration for wild-type virus) in all subjects, and 8OH-efavirenz concentration in CSF was >3.3 ng/mL (a proposed toxicity threshold) in 11 of 14 and 7 of 14 subjects randomized to the 400 mg and 600 mg doses of efavirenz, respectively. Whereas CSF efavirenz concentration was significantly associated with plasma concentration (P <. 001) and cytochrome P450 2B6 genotype (CSF efavirenz GG to GT/TT geometric mean ratio, 0.56 [90% CI,. 42-.74]), CSF 8OH-efavirenz concentration was not (P =. 242 for association and CSF 8OH-efavirenz GG to GT/TT geometric mean ratio, 1.52 [90% CI,. 97-2.36]). Conclusions. With both doses of efavirenz studied, CSF concentrations were considered adequate to inhibit HIV replication, although concentrations of 8OH-efavirenz were greater than those reportedly associated with neuronal toxicity. CSF exposure of 8OH-efavirenz was not dependent on plasma exposure and, as we postulate, may be subject to saturable pharmacokinetic effects.