Background: Few large projects have evaluated the factors that influence the HIV RNA concentrations (viral load) in cerebrospinal fluid (CSF) during antiretroviral therapy (ART) over time. We aimed to determine the correlates of HIV RNA in CSF in a large cohort. Methods: We analysed longitudinal data from adults living with HIV in the US CHARTER cohort. Participants in the CHARTER study were recruited from six US academic medical centres—in Baltimore (MD), Galveston (TX), New York (NY), St Louis (MO), San Diego (C92A), and Seattle (WA). Participants in this study had been assessed at least three times between Sept 4, 2003, and Sept 14, 2010, and were taking ART and underwent venous and lumbar puncture with measurement of HIV RNA concentration at all assessments. The lower limit of quantification of the HIV RNA assays was 50 copies per mL. Data were analysed with longitudinal mixed effects logistic regression to identify correlates of HIV RNA concentration (as a binary [detectable or not] and as a continuous variable) in CSF over time. We tested demographic characteristics, plasma HIV RNA, nadir and current CD4 cell count in blood, current CD8 cell count in blood, estimated duration of HIV infection, AIDS diagnosis, duration of ART, adherence to ART, ART characteristics, and CSF characteristics as potential correlates. Findings: At the time of analysis, 2207 assessments from 401 participants met the criteria for inclusion in this study. Mean duration of observation was 33·7 months (range 12–84). HIV RNA concentrations in 710 (32·2%) plasma specimens and in 255 (11·6%) CSF specimens were greater than the lower limit of quantification. The best multivariate model of HIV RNA concentration in CSF greater than the lower limit of quantification over time included increased plasma HIV RNA concentration (odds ratio 18·0 per 1 log10 copy per mL, 95% CI 11·3 to 28·8; p<0·0001), increased CSF leucocyte count (2·01 per 5 cells per μL, 1·61 to 2·39; p<0·0001), decreased CD4 cell count (0·53 per 5 square-root cells per μL, 0·35 to 0·79; p=0·0025), decreased CNS penetration-effectiveness value (0·71 per unit, 0·56 to 0·92; p=0·0078), increased CD8 cell count (1·51 per 5 square-root cells, 1·11 to 2·06; p=0·0089), and protease inhibitor use (3·26, 1·04 to 10·23; p=0·039; model R2=0·22, p<0·0001). Analyses of continuous HIV RNA concentration in CSF that accounted for censoring below the lower limit of quantification had similar findings, although increased HIV RNA concentrations in CSF were also associated with black ethnicity (change in log10 HIV RNA concentration in CSF 0·205, 0·0367 to 0·3733; p=0·017), increased total protein in CSF (0·0025, −0·0002 to 0·0052; p=0·069), and the presence of addictive-drug metabolites in urine (0·103, −0·013 to 0·219; p=0·081). Interpretation: The identified correlates of HIV RNA concentration in CSF during ART could strengthen clinical prediction of risk for failure to achieve or maintain HIV RNA suppression in CSF. Because most participants in this analysis were ART-experienced and were taking a three-drug regimen that did not include an integrase inhibitor, future research should focus on participants who are taking their first ART regimens or regimens that include integrase inhibitors or two drugs. Funding: The work was supported by the National Institute of Mental Health and the National Institute of Neurological Disorders and Stroke.