OBJECTIVE: To evaluate inter-reader agreement and diagnostic accuracy of chest radiography (CXR) in the diagnosis of tuberculosis (TB) in children with human immunodeficiency virus (HIV) infection. DESIGN: HIV-infected children with clinically suspected TB were enrolled in a prospective study conducted in Burkina Faso, Cambodia, Cameroon and Viet Nam from April 2010 to December 2014. Three readers-a local radiologist, a paediatric pulmonologist and a paediatric radiologist-independently reviewed the CXRs. Inter-reader agreement was then assessed using the kappa coefficient. Diagnostic accuracy of CXR was assessed in culture-confirmed cases and controls. RESULTS: A total of 403 children (median age 7.3 years, interquartile range 3.5-9.7; 49.6% males) were enrolled. Inter-reader agreement was as follows: between local radiologist and paediatric pulmonologist, kappa = 0.36 (95%CI 0.27-0.45); local radiologist and paediatric radiologist, kappa = 0.16 (95%CI 0.08-0.24); and paediatric pulmonologist and paediatric radiologist, kappa = 0.30 (95%CI 0.21-0.40). Among 51 cases and 151 controls, after a consensus, CXR had a sensitivity of 71.4% (95%CI 58.8-84.1) and a specificity of 50.0% (95%CI 41.9-58.1). Alveolar opacities and enlarged lymph nodes on CXR had limited specificity for TB (64.7% and 70.2%, respectively). Miliary and/or nodular opacities patterns on CXR were more specific to TB (specificity 94.3%). CONCLUSION: CXR showed poor-to-fair inter-reader agreement and limited diagnostic accuracy for TB in HIV-infected children, likely due to comorbidities. Radiological criteria for this specific population require further investigation.< Réduire