The performance of 10 pharmacokinetic models in predicting blood propofol concentrations was evaluated in patients during neurosurgical anesthesia. Eight patients-ASA category I or II, aged 49 +/- 18-years, weighing 71 +/- 20 kg, and scheduled for routine neurosurgery-were anesthetized with propofol and sufentanil using Ohmeda pumps controlled with a personal computer. Sufentanil was administered as a bolus of 0.3 microgram.kg-1, 5 min before induction of anesthesia, and infused at a constant rate of 0.5 microgram.kg-1.h-1 throughout the study. At induction, propofol was administered as a bolus of 1.5 mg.kg-1 followed by a continuous infusion of 6 mg.kg-1.h-1. During surgery, the propofol infusion rate was deliberately increased by 2 mg.kg-1.h-1 every 15 min up to 12 mg.kg-1.h-1. Arterial blood samples were drawn at the end of each infusion step for measurement of propofol concentrations by high-performance liquid chromatography. Measured propofol concentrations were compared with theoretical concentrations derived from 10 published pharmacokinetic models designed in different clinical settings. Each model has been assessed by calculating the median of the performance error, the median of the absolute performance error, and their 10th and 90th percentiles. Models designed for certain categories, such as children, young, or elderly patients who received propofol as a bolus injection, showed a bad predictive accuracy. The models of Gepts et al. (Anesth Analg 1987; 66:1256-1263, Anaesthesia 1988; 43(suppl):8-13), Tackley et al. (Br J Anaesth 1989;62:46-53), and Cockshott (Postgrad Med J 1985;61:55), derived from healthy patients receiving continuous propofol infusions, provided the best agreement between expected and measured propofol concentrations; they showed bias and inaccuracy lower than 17%. In conclusion, the accurate prediction of blood propofol concentrations from different continuous infusion rates in ASA I or II patients requires the selection of appropriate pharmacokinetic models derived from similar categories of patients and using a similar technique of propofol administration. However, in clinical practice, the selection of a specific set among the appropriate models is balanced by the interindividual variability in blood propofol concentrations adjusted to clinical effects.