An enzyme-linked immunosorbent assay (ELISA)-based investigation of anti-lipoarabinomannan (LAM) antibody levels in the sera of patients with acid-fast bacilli (AFB)-positive pulmonary tuberculosis (PTB), AFB-negative PTB and non-TB respiratory tract symptoms was conducted. The anti-LAM results were further evaluated using urine LAM detection and a clinical diagnostic score (DS) system as references. Using sputum AFB as a reference, positive anti-LAM was found in 66.9% of 139 AFB-positive PTB, 34.4% of 61 AFB-negative PTB and 23.5% of 800 non-TB patients and in 8% of 50 healthy individuals. The positive and negative predictive values were 48.7% and 87.4%, respectively. Using the DS as a reference, the sensitivity and specificity were 50.5% and 78.3%, respectively, whereas 45.8% of urine LAM positives and 77.9% of urine LAM negatives were correctly identified by the anti-LAM ELISA. In TB endemic areas a negative anti-LAM could be of practical value, particularly when other indicators of PTB are negative. Using any of these methods as a reference, a positive anti-LAM would mislead in about one-quarter of cases. Had all the 3 methods been combined and at least 2 positive tests sufficed, 90.6% of AFB-positive PTB, 52.5% of AFB-negative PTB and 94.9% of non-TB patients would have been correctly diagnosed. Apart from the possible impact of HIV, the low accuracy of the current assay could be due to intravascular formation of LAM-anti-LAM complexes, latent TB or environmental mycobacterial infections.