New tools to monitor treatment response and predict outcome from tuberculous meningitis (TBM) are urgently required. We retrospectively evaluated the 16S rRNA-based molecular bacterial load assay (MBLA) to quantify viable Mycobacterium tuberculosis in serial cerebrospinal fluid (CSF) from adults with TBM. 187 CSF samples were collected before and during the first two months of treatment from 99 adults TBM, comprising 56 definite, 43 probable or possible TBM, and 18 non-TBM and preserved at −80°C prior to MBLA. We compared MBLA against MGIT culture, GeneXpert MTB/RIF (Xpert) and Ziehl-Neelsen (ZN) smear. Before treatment, MBLA was positive in 34/99 (34.3%), significantly lower than MGIT 47/99 (47.5%), Xpert 51/99 (51.5%) and ZN smear 55/99 (55.5%). After one month of treatment, MBLA and MGIT were positive in 3/38 (7.9%) and 4/38 (10.5%), respectively, whereas Xpert and ZN smear remained positive in 19/38 (50.0%) and 18/38 (47.4%). In summary, MBLA was less likely to detect CSF bacteria before the start of treatment compared with MGIT culture, Xpert and ZN smear. MBLA and MGIT positivity fell during treatment because of detecting only viable bacteria, whereas Xpert and ZN smear remained positive for longer because of detecting both live and dead bacteria. Sample storage and processing may have reduced MBLA-detectable viable bacteria; and sampling earlier in treatment may yield more useful results. Prospective studies with CSF sampling after 1–2 weeks are warranted.