Research on the determinants of condom use and condom non-use generally has relied on self-reported data with questionable validity. We identified predictors of recent, unprotected sex among 331 female sex workers in Madagascar using two outcome measures: self-reports of unprotected sex within the past 48 h and detection of prostate-specific antigen (PSA), a biological marker of recent semen exposure. Multivariable logistic regression revealed that self-reported unprotected sex was associated with three factors: younger age, having a sipa (emotional partner) in the prior seven days, and no current use of hormonal contraception. The sole factor related to having PSA detected was prevalent chlamydial infection (adjusted odds ratio, 4.5; 95% confidence interval, 2.0-10.1). Differences in predictors identified suggest that determinants of unprotected sex, based on self-reported behaviors, might not correlate well with risk of semen exposure. Caution must be taken when interpreting self-reported sexual behavior measures or when adjusting for them in analyses evaluating interventions for the prevention of HIV/STIs.