Cancer patients have an increased risk of reactivation of latent tuberculosis infection. It is unknown which strategy on screening should be used in this population in developing countries. We aimed to determine the concordance between the tuberculin skin test (TST) and QuantiFERON®-TB (QFT) assay in order to diagnose latent tuberculosis infection in cancer patients. We conducted a cross-sectional study of the agreement of diagnostic tests. Prevalence and agreement between tests were calculated. A logistic regression to assess predictors of discordance was performed. The accuracy of the TST to predict QFT results by a receiver operating characteristic (ROC) curve was evaluated. We included 149 adults with cancer without active tuberculosis. Prevalence of latent tuberculosis infection was 21.5% (n=32), defined as positive results on either test. Test agreement was moderate for the diagnosis of latent tuberculosis infection (κ=0.43, 90% CI 0.26-0.6). No predictor was associated with the chance of discordant results. Agreement improved slightly using a cut-off point ⩾8 mm (κ=0.5, 90% CI 0.35-0.66). In a moderate-incidence setting, a moderate agreement was found between tests in cancer patients. Modification of the cut-off points of test results achieved marginally better agreement between the TST and QFT.