The pharmacokinetics and pharmacodynamics of midazolam and diazepam were compared after intravenous infusions of 0.03 and 0.07 mg/kg midazolam and 0.1 and 0.2 mg/kg diazepam on four separate occasions in 12 healthy male subjects in a randomized four-way crossover design. The Digit Symbol Substitution Test (DSST) was used as a measure of drug effect. Subjects performed three practice tests before dosing to account for any effects caused by familiarization ("learning curve") with the testing procedure. Pharmacokinetic and pharmacodynamic data were simultaneously fitted to a semiparametric model. In this model, a pharmacokinetic model related dose to plasma concentrations, a link model related plasma concentrations to the concentration at the effect site, and a pharmacodynamic model related the effect site concentration to the observed effect. The plasma-effect site equilibrium half-life was approximately 2 1/2 times longer for midazolam than for diazepam, which is in good agreement with previously published data. Based on the estimated effect site concentration at which half of the maximal effect was reached, midazolam had approximately a sixfold greater intrinsic potency than diazepam. This difference in potency was also observed in a previous study that used transformed electroencephalographic (EEG) data to assess pharmacodynamic activity. The findings reported here with a clinically relevant pharmacodynamic marker (DSST) confirm the utility of surrogate drug effect measures such as EEG. This work also shows the feasibility of conducting pharmacokinetic pharmacodynamic analysis during the drug development process.