OBJECTIVE: To evaluate the diagnostic performance of TB-LAMP, a manual molecular tuberculosis (TB) detection method, and provide comparison to the Xpert MTB/RIF assay. METHODS: In a large multicentre study, two sputum samples were collected from participants with TB symptoms in reference laboratories in Peru, South Africa, Brazil, and Vietnam. Each sample was tested with TB-LAMP. The reference standard consisted of four direct smears, four cultures, and clinical and radiological findings. Individuals negative on conventional tests were followed up after 8 weeks. The Xpert MTB/RIF assay was performed on fresh or frozen samples as a molecular test comparison. RESULTS: A total of 1036 adults with suspected TB were enrolled. Among 375 culture-confirmed TB cases with 750 sputum samples, TB-LAMP detected 75.6% (95% confidence interval (CI) 71.8-79.4%), including 97.9% (95% CI 96.4-99.4%) of smear-positive TB samples and 46.6% (95% CI 40.6-52.7%) of smear-negative TB samples. Specificity in 477 culture-negative participants not treated for TB (954 sputum samples) was 98.7% (95% CI 97.9-99.6%). TB-LAMP test results were indeterminate in 0.3% of cases. CONCLUSIONS: TB-LAMP detects nearly all smear-positive and half of smear-negative TB cases and has a high specificity when performed in reference laboratories. Performance was similar to the Xpert MTB/RIF assay.