Fentanyl was administered to 21 patients using a computer-controlled infusion pump (CCIP) based on a pharmacokinetic model. Eleven of the patients were doted according to the pharmacokinetics described by McClain and Hug, and ten of the patients were dosed according to the pharmacokinetics described by Scott and Stanski. The authors measured the difference between the measured arterial fentanyl concentrations and the concentrations predicted by the CCIP for each pharmacokinetic parameter set. The median absolute performance error (MDAPE) in patients dosed according to McClain and Hug's parameters was 61%, and the MDAPE in patients dosed according to Scott and Stanski's parameters was 33%. The population pharmacokinetics in these 21 patients were analysed using a pooled data technique. The pharmacokinetics of fentanyl in this population showed a smaller central compartment volume and a more rapid initial distribution half-life than previously estimated for fentanyl. The derived pharmacokinetic parameters described these patients well and also predicted the observed fentanyl concentrations from four previously published fentanyl studies with reasonable accuracy. Comparison of the parameters used by the authors with those of McClain and Hug demonstrated that dosing regimens designed from pharmacokinetic models can be fairly accurate at the times sampled in the original study but may not be accurate at time points not sampled in the original research. The authors concluded that although the pharmacokinetics of fentanyl administered by CCIP are the same as the pharmacokinetics of fentanyl administered by a bolus or constant rate infusion, a pharmacokinetic study using a CCIP may be particularly effective at characterizing the most rapid distribution pharmacokinetic parameters, and thus may provide parameters appropriate for subsequent use in a CCIP.