Valproate serum concentrations between 45 and 125 microg/mL are associated with the drug's efficacy in acute mania. Adaptive control dosing of valproate has not been fully studied in psychiatry. The objective of this study was to derive population pharmacokinetic (PK) parameters for valproate in healthy volunteers and to test the ability of these PK parameters to estimate concentrations in adult psychiatric patients using a Bayesian program. Population PK parameters for oral valproate were estimated from 18 PK studies in six healthy volunteers (1) using NPEM2. A Bayesian PK program using these population parameters was used to predict valproate concentration-time points in a second cohort of 21 adult psychiatry patients using 0, 1, or 2 prior concentrations. Estimated population parameters (mean +/- SD) were: Ka, 1.15+/-1.75/h; V, 0.14+/-0.042 L/Kg; and CL, 0.902+/-0.133 L/h. Bayesian valproate estimations using these parameters were negatively biased (underestimations) using zero prior concentration and unbiased using 1 or 2 prior concentrations. Mean error values (95% CI) in microg/mL for predictions using 0, 1, or 2 prior concentration-time points were -12.0 (-22.5, -1.5), -9.5 (-19.1, 0.1), and -2.5 (-11.1, 6.1), respectively, and mean absolute error values in microg/mL (95% CI) were 19.8 (12.6, 27.1), 16.3 (9.4, 23.3), and 10.1 (4.9, 15.2), respectively. Population parameters derived from healthy adult volunteers provided biased predictions of valproate concentrations in adult psychiatric patients. However, estimates using 1 or 2 valproate concentration time points predicted future concentrations that were precise and unbiased, given the wide therapeutic target range.