We compared the predictive performance of the existing Diprifusor and Schnider models, used for target-controlled infusion of propofol, with a new modification of the Diprifusor model (White) incorporating age and sex covariates. The bias and inaccuracy (precision) of each model were determined using computer simulation to replicate the infusion profiles in an earlier study of 41 patients undergoing surgery with propofol administered by target-controlled infusion and in which timed, measured blood propofol concentrations were available. Bias with the White model (5%) was significantly less (p < 0.0001) than with the Diprifusor (16%) or Schnider (15%) models. The White model showed a significant improvement in accuracy (inaccuracy 19%, p < 0.0001) relative to the Diprifusor (26%). Temporal changes were such that with all three models, bias differed at induction and recovery. None of the models accounted fully for the extent of inter-individual variation in propofol clearance, but the improved performance with the White model suggests it has merit.