Summary. It is unclear whether the current threshold for ‘high’ hepatitis C virus (HCV) RNA level (800 000 IU/mL) is optimal for predicting sustained virological response (SVR). We retrospectively analysed pretreatment HCV RNA levels and SVR rates in 1529 mono-infected and 176 HIV–HCV co-infected patients treated with peginterferon alfa-2a (40 kD) plus ribavirin. We improved the threshold for differentiating low and high viral load by fitting semiparametric generalized additive logistic regression models to the data and constructing receiver operating characteristics curves. Among HCV genotype 1 mono-infected patients, the difference in SVR rates between those with low and high baseline HCV RNA levels was 27% (70%vs 43%) when 400 000 IU/mL was used and 16% (59%vs 43%) when 800 000 IU/mL was used. In HIV–HCV genotype 1 co-infected patients, the difference was 51% (71%vs 20%) when 400 000 IU/mL was used and 43% (61%vs 18%) when 800 000 IU/mL was used. A lower threshold (200 000 IU/mL) was identified for genotype 1 mono-infected patients with ‘normal’ alanine aminotransferase (ALT) levels. No threshold could be identified in HCV genotype 2 or 3 patients. A threshold HCV RNA level of 400 000 IU/mL is optimal for differentiating high and low probability of SVR in genotype 1-infected individuals with elevated ALT.