Development of a non-sputum test using readily-obtainable biospecimens remains a global priority for tuberculosis (TB) control. We quantified lipoarabinomannan (LAM) concentrations, a pathogen biomarker for Mycobacterium tuberculosis, in urine, plasma and serum for real-world diagnostic accuracy of pulmonary TB among people living with and without HIV. We conducted a prospective diagnostic study among adults with TB symptoms in South Africa. We measured LAM concentrations in time-matched urine, plasma and serum with an electrochemiluminescence immunoassay using two capture antibodies (FIND 28 and S4-20). From the completed cohort, we randomly selected 210 participants (2 cases: 1 control) based on sensitivity estimates, and we compared diagnostic accuracy of LAM measurements against the microbiological reference standard. Urine and blood specimens from 210 of 684 adults enrolled were tested for LAM. Among 138TB-positive adults (41% female), median urine LAM was 137pg/mL and 52pg/mL by FIND 28 and S4-20, respectively. Average LAM concentrations were highest in HIV-positive participants with CD4+ T cells <200cells/mm3. Urine LAM by S4-20 achieved diagnostic sensitivity of 62% (95% CI: 53%-70%) and specificity of 99% (95% CI: 96%-100%). Plasma and serum LAM by FIND 28 showed similar sensitivity (70%, 95% CI: 62%-78%) and comparable specificities (90%, 95% CI: 82%-97%; 94%, 95% CI: 88%-99%). Diagnostic sensitivity of urine LAM by S4-20 was higher among participants without HIV (41%, 95% CI: 24%-61%) compared to HIV-positive participants with CD4 ≥200cells/mm3 (20%, 95% CI: 8%-39%). Detection of LAM was achievable in non-sputum specimens for pulmonary TB, but additional analyte concentration or signal amplification may be required to achieve diagnostic accuracy targets. Bill and Melinda Gates Foundation.