Abstract Background Timely diagnosis of tuberculous meningitis (TBM) remains challenging. Molecular diagnostic tools are necessary, particularly in low- and middle-income countries. There is no approved commercial polymerase chain reaction (PCR) assay that can be used to detect Mycobacterium tuberculosis in non-respiratory samples, such as the cerebrospinal fluid (CSF). We aimed to validate the threshold cycle (Ct) cut-off points; calculate the operational characteristics of real-time PCR for detection of M. tuberculosis (MTb qPCR) in the CSF; and the inhibitory affect of CSF red blood cells (RBC) and total proteins on MTb qPCR. Methods A total of 334 consecutive participants were enrolled. Based on clinical, laboratory and imaging data, cases of suspected TBM were categorized as definite, probable, possible or not TBM cases. Receiver operating characteristic curve analysis was used to select the best discriminating Ct value. Results For TBM cases categorized as definite or probable (n=21), the Ct validated for CSF (≤39.5) improved the diagnostic performance of MTb qPCR on CSF samples. The sensitivity was 29%, specificity was 95%, positive predictive value was 26%, negative predictive value was 95%, efficiency was 90% and positive likelihood was 5.3. The CSF RBC and total protein did not affect the positivity of the MTb qPCR. Conclusions These data support the validation of a highly specific but low sensitive MTb qPCR assay for the TBM diagnosis using CSF samples. MTb qPCR contributes significantly to the diagnosis, mainly when associated with conventional microbiology tests and clinical algorithms.