Direct capture enzyme-linked immunosorbent assay (ELISA) for lipoarabinomannan (LAM) was performed on urine samples from 200 tuberculosis (TB) patients and 800 non-TB patients routinely diagnosed among consecutive suspects in an Ethiopian TB centre. 50 healthy Ethiopians, 50 healthy individuals and 100 non-TB patients from Norway served as controls. Of the TB patients, 139 (69.5%) were positive for acid-fast bacilli (AFB). In the remaining cases the diagnosis was based on suggestive clinical findings. All Ethiopian non-TB patients were AFB negative and showed no clinical evidence of TB. In the Ethiopian groups, 148 (74%) of the TB patients, 105 (13.1%) of the non-TB patients and 5 (10%) of the healthy controls were positive by the LAM-ELISA. 113 (81.3%) of AFB positives and 35 (57.4%) of AFB-negative TB patients had positive LAM-ELISA. In the Norwegian groups all were LAM negative. The sensitivity and specificity of the LAM-ELISA for TB patients versus Ethiopian non-TB patients were 74% and 86.9%, respectively; the positive and negative predictive values were 58.5% and 93.0%. This study suggests that detection of LAM in the urine of TB patients may improve case finding and that diagnostic tests based on this principle may serve as valuable supplemental tools in TB control.