The objective was to develop a population pharmacokinetic-pharmacodynamic model of caffeine's psychomotor effects in healthy, non-habitual users of caffeine. Twenty Chinese males each received a single dose of 250 mg of caffeine orally. Plasma concentrations of caffeine were determined at various times within 24 h after dosing. The subjects' psychomotor performance was evaluated before and at various times after dosing by a test battery consisting of oculomotor assessment (saccadic velocity) as well as the computerised Swedish Performance Evaluation System. Nonlinear mixed-effects modelling to analyse the pharmacokinetic-pharmacodynamic relationships was performed using NONMEM. Model robustness was assessed by a nonparametric bootstrap. The results showed that caffeine caused significant improvements in psychomotor functioning. The time course of these effects was best described by pharmacokinetic/pharmacodynamic models involving an effect compartment. The transfer half-lives between plasma and effect site for different domains of psychomotor functioning were in the range 24.8-49.5 min. Evaluation of the final models showed close agreement between pairs of bootstrapped and final model parameter estimates (all differences<10%). These results provided the first suggestive evidence that caffeine effects on psychomotor performance occur after some time delay relative to changes in plasma caffeine concentration. The models for the neurobehavioural tests provided similar transfer half-lives between plasma and effect site.