Background: Urine tests for mycobacterial lipoarabinomannan might be useful for point-of-care diagnosis of tuberculosis in adults with advanced HIV infection, but have not been assessed in children. We assessed the accuracy of urine lipoarabinomannan testing for the diagnosis of pulmonary tuberculosis in HIV-positive and HIV-negative children. Methods: We prospectively recruited children (aged ≤15 years) who presented with suspected tuberculosis at a primary health-care clinic and paediatric referral hospital in South Africa, between March 1, 2009, and April 30, 2012. We assessed the diagnostic accuracy of urine lipoarabinomannan testing with lateral flow assay and ELISA, with mycobacterial culture of two induced sputum samples as the reference standard. Positive cultures were identified by acid-fast staining and tested to confirm Mycobacterium tuberculosis and establish susceptibility to rifampicin and isoniazid. Findings: 535 children (median age 42·5 months, IQR 19·1-66·3) had urine and two induced specimens available for testing. 89 (17%) had culture-confirmed tuberculosis and 106 (20%) had HIV. The lateral flow lipoarabinomannan test showed poor accuracy against the reference standard, with sensitivity of 48·3% (95% CI 37·6-59·2), specificity of 60·8% (56·1-65·3), and an area under the receiver operating characteristic curve of 0·53 (0·46-0·60) for children without HIV and 0·64 (0·51-0·76) for children with HIV. ELISA had poor sensitivity in children without HIV (sensitivity 3·0%, 95% CI 0·4-10·5) and children with HIV (0%, 0·0-14·3); overall specificity was 95·7% (93·4-97·4). Interpretation: Urine lipoarabinomannan tests have insufficient sensitivity and specificity to diagnose HIV-positive and HIV-negative children with tuberculosis and should not be used in this patient population.