The molecular analysis of cerebrospinal fluid (CSF) provides an inestimable tool for the study of HIV infection of the central nervous system (CNS). Current nucleic acid amplification techniques enable the measurement of CSF HIV-1 RNA levels which can be predictive of HIV-associated neurological damage. CSF HIV-1 RNA levels do not necessarily correlate with the corresponding plasma levels, thus supporting the possibility of an intrathecal virus production, i.e., from brain macrophages. However, in early stages of HIV infection, as well as during some opportunistic CNS diseases, CNS or CSF infiltrating lymphocytes might be the main source of CSF virus. A drastic decrease in CSF viral load is usually observed along with a decrease in plasma levels in patients receiving highly active antiretroviral therapy (HAART), with durable suppression of CSF viral load over months. However, during the first weeks of therapy, the dynamics of response may differ in the CSF as compared to plasma, again suggesting that virus replication may be compartmentalised in the CSF. A number of mechanisms are likely to be involved in the response to therapy in CSF, including among the others the trafficking of cell populations supporting viral replication between blood, CNS and CSF, and the role of the anatomical brain barriers in limiting the access of antiretroviral drugs into the CSF. A potential risk associated with compartmentalisation of HIV infection is of an incomplete suppression of virus replication in the CSF, thus creating the ground for local development of anti-HIV drug resistance. In order to assess this occurrence, long-term studies of viral load and genotypic analyses on paired CSF and plasma will be necessary and these will also help elucidate the complex interrelationship between viral replication in these compartments.