BACKGROUND: Liver toxicity due to tuberculosis (TB) treatment is a frequent cause of treatment interruption, and may sometimes lead to a change in therapy to a less potent regimen. OBJECTIVE: To estimate the risk of hepatotoxicity in patients with or without hepatitis B virus (HBV) infection receiving TB treatment and to develop a clinical prediction rule. DESIGN: A prospective observational follow-up was conducted. Data from 154 patients who underwent TB treatment were analysed. Crude risk ratios were estimated and a Cox proportional hazards model was fit. RESULTS: The mean follow-up time was 187 days. Crude risk ratios showed that ethnicity, human immunodeficiency virus infection, multiple sexual partners, highly active antiretroviral treatment, and clinical forms of TB were possible predictors of liver toxicity. HBV infection and other sexually transmitted diseases showed considerable relative risk, although not statistically significant. The Cox proportional hazards model identifi ed the following predictors of hepatotoxicity: White ethnicity, multiple sexual partners, high baseline alanine transferase and clinical forms of TB. Active HBV, indicated by the detection of surface antigen HBV, could predict hepatotoxicity, although with low precision. CONCLUSION: Using this information, we were able to apply a score and draw a nomogram to estimate survival probabilities and median times to event for each patient.