BACKGROUND: Detection of the mycobacterial cell wall antigen lipoarabinomannan (LAM) in urine can be used to diagnose HIV-associated tuberculosis (TB) using a qualitative (positive/negative) read-out. However, it is not known whether the quantity of LAM present in urine provides additional prognostic information. METHODS/FINDINGS: Consecutively recruited adult outpatients initiating antiretroviral therapy (ART) in South Africa were investigated for TB regardless of clinical symptoms using sputum smear microscopy and liquid culture (reference standard). Urine samples were tested using the Clearview TB-ELISA for LAM and the Xpert MTB/RIF assay. The ELISA optical densities (OD) were used as a quantitative assessment of urine LAM. Among 514 patients with complete sputum and urine LAM OD results, culture-confirmed TB was diagnosed in 84 patients. Twenty-three (27.3%) were LAM-positive with a median LAM OD of 0.68 (IQR 0.16-2.43; range, 0.10-3.29) and 61 (72.6%) were LAM negative (LAM OD <0.1 above background). Higher LAM ODs were associated with a range of prognostic indices, including lower CD4 cell counts, lower haemoglobin levels, higher blood neutrophil counts and higher mycobacterial load as assessed using both sputum and urine samples. The median LAM OD among patients who died was more than 6.8-fold higher than that of patients who remained alive at 3 months (P<0.001). The small number of deaths, however, precluded adequate assessment of mortality risk stratified according to urine LAM OD. CONCLUSIONS: In patients with HIV-associated TB, concentrations of LAM in urine were strongly associated with a range of poor prognostic characteristics known to be associated with mortality risk. Urine LAM assays with a semi-quantitative (negative vs. low-positive vs. high-positive) read-out may have improved clinical utility over assays with a simple binary result.