OBJECTIVE: To assess the performance of WHO clinical and CD4 cell count criteria for antiretroviral treatment (ART) failure among HIV-infected adults in a workplace HIV care programme in South Africa. DESIGN: Cohort study. METHODS: We included initially ART-naive participants who remained on first-line therapy and had an evaluable HIV viral load result at the 12-month visit. WHO-defined clinical and CD4 cell count criteria for ART failure were compared against a gold standard of virological failure. RESULTS: Among 324 individuals (97.5% men, median age 40.2, median starting CD4 cell count and viral load 154 cells/mul and 47,503 copies/ml, respectively), 33 (10.2%) had definite or probable virological failure at 12 months, compared with 19 (6.0%) and 40 (12.5%) with WHO-defined CD4 and clinical failure, respectively. CD4 criteria had a sensitivity of 21.2% and a specificity of 95.8% in detecting virological failure, and clinical criteria had sensitivity of 15.2% and specificity of 88.1%. The positive predictive value of CD4 and clinical criteria in detecting virological failure were 36.8 and 12.8%, respectively. Exclusion of weight loss or tuberculosis failed to improve the performance of clinical criteria. CONCLUSION: WHO clinical and CD4 criteria have poor sensitivity and specificity in detecting virological failure. The low specificities and positive predictive values mean that individuals with adequate virological suppression risk being incorrectly classified as having treatment failure and unnecessarily switched to second-line therapy. Virological failure should be confirmed before switching to second-line therapy.